Rapid optimization of the hydrolysis of N-trifluoroacetyl-S-tert-leucine-N-methylamide using high-throughput chemical development techniques

Article abstract of DOI:10.1021/op010011s

Our efforts are focused on the application of automation to Process R&D. This article will describe the application of high throughput methods to rapidly investigate a development challenge. In this case we needed to study the deprotection of N-trifluoroacetyl-S-tert-leucine-N-methylamide which afforded a lower than expected yield when subjected to standard deprotection reaction conditions. This chemistry was systematically investigated by a sequential series of high-throughput experiments using various automated and semi-automated systems. The studies included a combinatorial screen of discrete reaction conditions, a screening DOE to study a broad range of continuous factors, and a two-factor central composite design to optimize the important factors. By applying high-throughput methods we were able to optimize the yield of the reaction by performing a large number of experiments in a short period of time.

Full text of DOI:10.1021/op010011s

Organic Process Research & Development 2001, 5, 294−298
Rapid Optimization of the Hydrolysis of
N′-Trifluoroacetyl-S-tert-leucine-N-methylamide Using High-Throughput
Chemical Development Techniques
Victor W. Rosso,* James L. Pazdan, and John J. Venit
Process Research and DeVelopment, Pharmaceutical Research Institute, Bristol-Myers Squibb Company,
New Brunswick, New Jersey 08903, U.S.A.
Abstract:
Our efforts are focused on the application of automation to
Process R&D. This article will describe the application of high
throughput methods to rapidly investigate a development
Figure 1. Hydrolysis of trifluoroacetyl-L-tert-leucine-N-methyl
challenge. In this case we needed to study the deprotection of
N′-trifluoroacetyl-S-tert-leucine-N-methylamide which afforded
a lower than expected yield when subjected to standard
deprotection reaction conditions. This chemistry was systemati-
cally investigated by a sequential series of high-throughput
experiments using various automated and semi-automated
systems. The studies included a combinatorial screen of discrete
reaction conditions, a screening DOE to study a broad range
of continuous factors, and a two-factor central composite design
to optimize the important factors. By applying high-throughput
methods we were able to optimize the yield of the reaction by
performing a large number of experiments in a short period of
time.
amide.
We generally use three types of designed experiments to
accomplish this task. Combinatorial screens of discrete
factors are used to identify the best performing reagent/
solvent pairs for the chemical transformation. The screening
DOE technique uses fractional factorial designs to assess a
large number of discrete and continuous factors to identify
which factors have the greatest impact on the desired results.
Since the screening DOE measures multiple responses (e.g.,
yield) versus multiple factors, the resulting multidimensional
analysis returns a massive amount of chemical knowledge
for the time and material invested. The optimization DOE
uses central composite designs to study the significant factors
to arrive at the optimal reaction conditions. We have had
many successful automated studies that use these techniques
to rapidly optimize reactions for scale up.
An example of the application of these techniques to the
chemistry in Figure 1 will be discussed. During the course
of a study of an alternative synthetic strategy, we encountered
a sluggish deprotection reaction that afforded lower than
anticipated yields. The application of known3 N-trifluoro-
acetyl deprotection conditions resulted in a slow reaction,
and the use of more forcing conditions resulted in low yields
(62-66%). Since the tert-leucine moiety accounted for much
of the cost of the molecule, it was essential that the yield
for this conversion be >90% for the route to be cost-
effective. Therefore, we commenced a rapid automated study
of this system.
Introduction
Process automation efforts have focused on the introduc-
tion of new tools and techniques to Process R&D to enhance
productivity throughout the pharmaceutical industry.1 This
includes not only the introduction of modular equipment such
as reactor blocks and liquid handlers to automate the
performance of experiments but also experimental strategies
such as design of experiments2 (DOE), and high-throughput
synthetic techniques that allow process chemists to take full
advantage of rapidly developing automation capabilities. One
such application is to use a combination of screening and
statistically designed experiments to simultaneously perform
rapid and thorough investigations of chemical processes.
These studies allow us to quickly develop the scientific
understanding necessary for the implementation of processes
at pilot scale.
The first step in our strategy for studying a chemical
reaction is to systematically list all potential variables for
the reaction. Since the typical chemical reaction contains both
quantitative (continuous) factors such as temperature, equiva-
lents of reagents, etc. and qualitative (discrete) factors such
as solvents, identities of reagents, etc., a strategy is required
to systematically study broad ranges of reaction conditions.
The study consisted of the following phases: (1) a
thorough listing of all reaction conditions, (2) a combinatorial
screen of all discrete factor combinations, (3) a multivariable
screening DOE of all the continuous factors, and (4) an
optimization DOE on important factors identified in previous
experiments. By applying this type of study, our objective
was not only to identify the optimal reaction conditions in a
short period of time but also to investigate the broadest range
of reaction conditions so as to avoid the need to redevelop
the reaction with “better” reagents in the future.
(1) Harre, M.; Tilstam, U.; Weinmann, H. Org. Process Res. DeV. 1999, 3,
304-318.
(2) Hicks, C. R.; Turner, K. V., Jr. Fundamental Concepts in the Design of
Experiments; Oxford University Press: New York, 1999.
(3) Schmidt, U.; Weinbrenner, S. Synthesis 1996, 1, 28-30. (b) Swaminathan,
S. Internal communication.
294
•
Vol. 5, No. 3, 2001 / Organic Process Research & Development
10.1021/op010011s CCC: $20.00 © 2001 American Chemical Society and The Royal Society of Chemistry
Published on Web 04/27/2001

Table 1. List of all potential variables for the reaction
reaction conditions (0.125 M) were used for the screen to
obtain a manageable volume for the small-scale reactions.
The reactions were performed in 50% aqueous methanol with
the exception of the stronger bases KOMe, KOtBu, and DBU
that were run in 100% primary solvent (0.125 M) instead of
50% aqueous to avoid decomposition of the base.
a
b
base
c
d
e
f
solvent
equiv of % water temp
base
concn
(mL/g (M))
water
methylamine
K2CO3
KHCO3
1
2
3
0
50
100
30
5 (0.84 M)
methanol
ethanol
THF
45 10 (0.42 M)
60 15 (0.28 M)
The reactions were performed by dispensing 100 µL of a
0.5 M stock solution of compound (2) in methylene chloride
into each reactor. The solvent was evaporated in a Savant
SpeedVac, and the appropriate reaction solvent (Table 1,
column a, 200 µL) was dispensed into each reactor. The
appropriate base (Table 1, column b, 2 equiv of, 200 µL of
0.5 M aqueous base) was then dispensed into each reaction
vessel. The bases KOMe, KOtBu, and DBU were added neat,
with an additional charge of the primary reaction solvent.
The reactors were incubated at 45 °C for 3 h on a J-Kem
shaker plate. The reactors were then cooled, and aliquots
were taken for HPLC analysis. Due to the higher variability
associated with quantitation of the small volumes used in
the screening study, the apparent purity of the product as
measured by HPLC was used to compare these experiments
(Table 2).
Analysis of the full combinatorial screen allows us to
compare mean values for each solvent and base in addition
to the data for each solvent/base pair. The mean data for the
solvents show that water is the best overall solvent for this
reaction (Figure 2). Analysis of the mean data for the various
bases as depicted in Figure 3 suggests that DBU is the best
reagent for this transformation. However, when the solvent/
base pair data is reviewed, we find the best overall combina-
tion is aqueous lithium hydroxide. This result is in agreement
with the mean result for solvents. The mean value for bases
does not agree, and this was caused by the poor performance
of lithium hydroxide in the presence of several organic
solvents. This is an example of strong two-way interactions
that can occur between chemicals, underscoring the impor-
tance of including as many combinations of factors as
possible in these screens to avoid missing unexpected results
that occur with certain unique combinations of reagents.
On the basis of the data generated by this screening study,
we decided to pursue aqueous hydroxide reactions for the
next round of experimentation based on the following
criteria: the strong performance of LiOH in water, the
environmental friendliness of aqueous reaction conditions,
ethanolamine
acetonitrile LiOH
ethyl acetate K2HPO4
2-propanol DBUa
KOMea
KOtBua
DiPEA
a Run without water as cosolvent.
Results and Discussion
Standard reaction conditions for this conversion were 1.5
equiv of sodium hydroxide in 50% aqueous methanol (0.3
M) at 45 °C for 2 h.3 Application of these conditions resulted
in only minimal conversion of starting material to product
(<10% yield). The use of more vigorous (60 vs 45 °C)
conditions drove the reaction to completion but resulted in
poorer than expected yields (63-66%). To identify a system
that would hydrolyze the N-trifluoroacetyl group in high yield
without appreciable amide hydrolysis, a list of potential
reaction conditions (Table 1) was assembled. Examination
of all combinations of these factor settings would require a
large number of reactions to be performed. Therefore we
studied these experimental conditions in groups of statisti-
cally designed experiments to more efficiently cover the
broad range of experimental conditions.
Screening Study
The first phase of this study was a combinatorial screen
of all of the solvent/base pairs outlined in Table 1 using a
general set of reaction conditions. The bases were selected
to give a broad sampling of different types and strengths of
base for the screen. For example, lithium hydroxide was
chosen to represent the group of hydroxides. Since the more
forcing conditions resulted in lower yields, we used the best
known reaction conditions (2 equiv of base, 50% aqueous
methanol, 45 °C) to study this array of reagents. More dilute
Table 2. HPLC area percent (AP) results of screening solvents vs bases
cosolvent
base
water THF acetonitrile ethyl acetate methanol ethanol 2-propanol mean
50% aq
50% aq
50% aq
50% aq
50% aq
50% aq
organic
organic
organic
50% aq
methylamine
34.1
14.2
54.1
12.5
82.8
15.2
48.6
72.8
47.6
12.5
4.0
3.1
26.2
0.8
9.7
1.0
2.5
54.8
2.7
5.2
8.6
52.8
1.5
8.3
x
x
x
x
0.7
14.7
3.6
4.0
8.5
3.1
1.7
4.9
38.6
3.4
0.4
10.8
4.0
24.3
2.9
3.4
4.7
3.8
49.0
4.1
3.8
2.8
30.8
1.7
70.9
1.9
3.1
52.3
3.3
0.7
5.6
1.6
2.8
2.0
2.1
0.6
2.1
56.6
2.9
0.6
11.2
5.4
27.9
4.3
25.8
4.2
10.8
54.0
10.7
2.3
potassium bicarbonate
potassium carbonate
ethanolamine
lithium hydride
potassium phosphate dibasic
potassium tert butoxide
DBU
potassium methoxide
DiPEA
0.7
0.7
mean
39.4
10.6
12.9
8.3
10.8
17.1
7.7
15.2
Vol. 5, No. 3, 2001 / Organic Process Research & Development
•
295

Figure 2. Average AP product by HPLC in various solvents.
Figure 3. Average AP product by HPLC in various bases.
and the low cost of aqueous hydroxides as opposed to that
of DBU.
amount of water and dilute aqueous hydroxide solution to
each reaction. After the reactions were heated and stirred
for 3 h, samples were taken for HPLC assay. The yields of
the reaction were then calculated based upon HPLC quan-
titation.
Screening DOE
The control experiments (2 equiv of base, 0.42 M, 45
°C) resulted in 97% conversion of starting material to product
and an average yield of 87%. Of the four factors, only
equivalents of base had a statistically significant impact on
conversion of starting material to product. With 1 equiv of
base, we had an average of 19% unreacted starting material,
whereas with 2-3 equiv of base, the reactions proceeded to
completion. These data suggested we needed to further
investigate the range between 1 and 2 equiv of base to
determine the necessary amount of base for this conversion.
With the selection of aqueous alkoxide solutions for the
screening DOE, we were left with the remaining variables
to study: the hydroxide for the reaction (lithium, sodium,
or potassium hydroxide), the overall concentration of the
reaction (0.28 to 0.84 M), reaction temperature (30 to 60
°C), and equivalents of base (1 to 3 equiv). We performed
a full factorial 23 experiment with each base, with an
additional six center points distributed among the three bases
to measure experimental error. The experiments were
performed by a Zymark Robot that dispensed the appropriate
296
•
Vol. 5, No. 3, 2001 / Organic Process Research & Development

Table 3. Yields for hydrolysis reaction under various
Table 5. Experimental Data for Optimization DOE
reaction conditions by HPLC
expt vol (mL water/g) concentration [M] equiv of base yield
equiv of base
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
15
20
15
20
10
0.28
0.21
0.28
0.21
0.42
0.65
0.42
0.28
0.21
0.21
0.28
0.42
0.18
0.28
0.42
0.28
1.925
1.75
1.5
1.25
1.25
1.5
1.75
1.5
1.75
1.25
1.5
1.75
1.5
1.5
1.25
1.075
88.3
89.6
92.0a
93.3
85.8
78.8
89.4
99.3a
97.8
98.9
95.6a
93.6
94.5
100.1a
86.8
95.0
1 equiv
2 equiv
0.42 M
3 equiv
temp,
base
°C
0.28 M 0.84 M
0.28 M 0.84 M
30
45
60
30
60.7
58.0
92.2
49.7
LiOH
92.3, 85.7
84.6, 88.9
87.1, 83.5
6.5
98.6
83.4
68.9
50.9
92.3
97.5
55.3
57.4
10
15
20
20
15
10
23.5
15
10
15
NaOH
KOH
45
60
30
97.2
78.1
58.4
51.7
98.8
95.5
56.8
50.4
45
60
90.1
53.7
105.8a
58.9
a Yields of >100% are caused by experimental error during sample preparation
(determined to be (4.2%).
a Replicate center points to assess experimental variability.
Table 4. Two-way interaction of equivalents vs
concentration
Table 6. Significance of the constants for the model equation
constant
parameter
p-value
significance
concentration
1 equiv
84.7
2 equiv
87.0
3 equiv
97.0
k2
k3
k4
k5
k6
vol
0.0039
0.6889
0.1725
0.0092
0.1415
high
0.28 M
0.42 M
0.84 M
equiv of base
vol × (equiv of base)
vol2
56.9
54.8
high
equiv of base2
The quantitated yields of the reactions with of various
factor settings are outlined in Table 3. The main effects that
each factor had on the yield of the reaction are detailed below
in order of importance:
• Concentration: most significant yield improvement
occurred with more dilute reactions.
mental parameters for concentration were expressed in terms
of mL of water/g of starting material for the purposes of the
design and analysis of this experiment. The equivalents of
base was centered at 1.5 to investigate the range of 1.075-
1.925 equiv of base. The reactions were prepared on a Gilson
215 liquid handler, heated for 3 h at 45 °C in a reactor block,
and sampled for HPLC analysis. Quantitative analysis of the
samples by HPLC affords yield data shown in Table 5.
A total of four replicates were performed at the center
points to measure experimental variability which is used to
determine if the observed effects are statistically significant.
The p-values assessing the statistical significance of each
potential term in the quadratic equation are listed in Table
6. Evaluation of the yield of product showed statistically
significant improvements caused by more dilute reaction
conditions, but no significant impact was observed for the
equivalents of base used. After removal of the insignificant
terms (p > 0.05) we obtain the following model for the yield
of the reaction:
• Temperature: higher temperatures afforded better yields.
• Equiv of base: there was an increase in yield when
excess base was used with more dilute reaction conditions.
• Identity of base: no significant effect was observed.
Of the experimental factors, the concentration was the
most significant factor affecting the yield. Additionally, there
was also a significant two-way interaction between the
concentration and the equivalents of base used (Table 4).
When 3 equiv of base was used, the concentration had a
greater influence on yield than when 1 equivalent of base
was used.
The conclusions from the screening DOE were: concen-
tration and equivalents of base would be the key factors for
an optimization DOE; the identity of the hydroxide was not
significant, justifying the selection of sodium hydroxide for
further optimization; and the temperature of the reaction did
not have a major impact upon yield (with dilute concentra-
tions, and excess base), and thus we selected the center point,
45 °C, for further optimization.
yield ) k1 + k2(vol) + k5(vol)2
The relationship between yield and volume is plotted in
Figure 4 and is given by the equation:
yield ) 54.74 + 4.877(vol) - 0.1381(vol)2
Optimization Designed Experiment
To optimize the reaction for equivalents of base and for
concentration, we used a two-factor central composite design
with four center points. This study was centered near the
best concentration (0.28 M) identified in the screening DOE
and was varied over a range of 0.17-0.65 M. The experi-
The optimization DOE shows good yield at water volumes
of 15 mL/g and greater with the model predicting a localized
maximum at 17.5 mL/g. The true optimal conditions would
be arrived at by factoring the cost of reagents versus the
Vol. 5, No. 3, 2001 / Organic Process Research & Development
•
297

Figure 4. Optimization DOE model of yield vs mL/g concentration.
savings of performing the reactions at lower volumes using
the information from this study. With respect to equivalents
of base, there were no significant effects detected in the range
investigated. Therefore, conditions we further studied were
0.24 M (17.5 mL/g starting material) in water with 1.25 equiv
of aqueous sodium hydroxide at 45 °C. A scale up of these
conditions resulted in an isolated yield of 95% of product
that is ready to be coupled in the next reaction in the synthetic
sequence.
chloride, Vilsmeier reagent, (28.7 g, 224 mmol, 1.2 equiv)
was then charged to the reaction while maintaining the
temperature below -15 °C for 2 h. The reaction slurry was
then quenched into excess aqueous methylamine (60 mL,
800 mmol, 4.3 equiv). The phases were separated, and the
organic phase was washed twice with water and concentrated
to minimal volume. The product was crystallized by the
addition of MTBE to afford 39.0 g (162 mmol, 87.4%).
HPLC AP 100, data: white solid, mp 160-162 °C, IR: KBr
3385, 2970, 1711, 1654, 1572, 1220, 1188, 1160, 738 cm-1;
1H NMR (CDCl3) δ 0.98 (s, 9H), 1.52 (s, NH2), 2.82 (S,
CH3), 3.15 (s, CH), 6.83 (Br, NH) ppm; EA Theory 45.00%
C, 6.29% H, 11.66% N, 23.73% F; Found 45.15% C, 6.45%
H, 11.63% N, 23.53% F.
S-tert-Leucine-N-methylamide. N′-Trifluoroacetyl-S-tert-
leucine-N-methylamide (1.66 g, 6.9 mmol) was charged to
a 50-mL reaction flask. Water (24.2 mL) and 10 N aqueous
NaOH (0.86 mL, 1.25 equiv) were charged, and the slurry
was heated to 45 C for 3 h. The solution was assayed to
contain 95.6 mol % S-tert-leucine-N-methylamide by in-
process HPLC versus a reference standard supplied by Great
Lakes Fine Chemicals.
Conclusions
We were able to rapidly and thoroughly optimize a typical
chemical reaction using combinatorial screens of discrete
variables, screening DOEs of multiple factors, and optimiza-
tion DOE. Important results from this study are the scientific
approach and automated execution of a study that covered a
broad range of reaction conditions in a systematic manner.
Specifically, we performed 116 reactions in 4 days to
optimize the yield from 63 to >95%. This study not only is
a model for what can be accomplished when these techniques
are applied to more challenging chemical reactions but also
helps us to define the capabilities of automated equipment
required to accelerate the pace of chemical process develop-
ment.
HPLC analysis was performed on a Shimadzu Discovery
VP Walk Up System. A volume of 10 µL was injected on a
YMC ODS-A S3µ 4.6 × 50 mm column. A gradient elution
from 10 to 50% CH3CN in 0.05 M aqueous NH4OAc over
4 min at a flow of 1.5 mL/minute and detection at λ 210.
Retention times: TFA-N-t-Leu-NHMe, 1.4 min; NH2-t-Leu-
NHMe, 0.5 min.
Methods and Materials
N′-Trifluoroacetyl-S-tert-leucine-N-methylamide. tert-
Leucine (23.96 g, 187 mmol) was charged to a 500-mL,
round-bottomed flask with methanol (48 mL), and 4.4 M
methanolic potassium methoxide (46 mL, 202.4 mmol, 1.1
equiv). After dissolution, ethyl trifluoroacetate (25 mL, 210
mmol 1.12 equiv) was charged, and the reaction mixture was
maintained at 40 °C for 2 h. The solution was then quenched
into 2 N aqueous HCl (113 mL, 226 mmol, 1.21 equiv).
Butyl acetate was charged (200 mL), and the phases were
separated. The organic phase was washed twice with water
(50 mL) and concentrated to an oil.
Acknowledgment
We gratefully acknowledge the suggestions and advice
of James Bergum and James Kenyon for the statistical design
of experiments and the work of A. Erik Rubin, Joseph Nolfo,
Barbara M. Ciaramella, and Susan D. Boettger in the design
of automated equipment.
Received for review February 6, 2001.
OP010011S
The oil was suspended in ethyl acetate (250 mL) and
cooled to -15 °C. (Chloromethylene)dimethylammonium
298
•
Vol. 5, No. 3, 2001 / Organic Process Research & Development