Transitioning enantioselective indicator displacement assays for α-amino acids to protocols amenable to high-throughput screening

Article abstract of DOI:10.1021/ja8038079

Enantioselective indicator displacement assays (eIDAs) for α-amino acids were conducted in a 96-well plate format to demonstrate the viability of the technique for the high-throughput screening (HTS) of enantiomeric excess (ee) values. Chiral receptors [Cu^II(1)]²⁺ and [Cu ^II(2)]²⁺ with the indicator chrome azurol S were implemented for the eIDAs. Enantiomeric excess calibration curves were made using both receptors and then used to analyze true test samples. These results were compared to those previously obtained with a conventional UV-vis spectrophotometer, and they showed little to no loss of accuracy, while the speed of analysis was increased. A sample of valine of unknown ee was synthesized through an asymmetric reaction to produce a realistic reaction sample, which was analyzed using receptor [Cu^II(1)]²⁺. The experimentally determined ee using our eIDA was compared to that obtained by chiral HPLC and ¹H NMR chiral shift reagent analysis. This gave errors of 4.7% and 12.0%, respectively. In addition to the use of ee calibration curves, an artificial neural network (ANN) was used to determine the % L-amino acid of the test samples and of the sample of valine of unknown ee from the asymmetric reaction. This method obtained errors of 5.9% and 2.2% compared to chiral HPLC and ¹H NMR chiral shift reagent analysis, respectively. The technique using calibration curves for the determination of ee on a 96-well plate allows one to determine 96 ee values in under a minute, enabling its use for HTS of asymmetric reactions with acceptable accuracy.

Full text of DOI:10.1021/ja8038079

Published on Web 08/21/2008
Transitioning Enantioselective Indicator Displacement Assays
for r-Amino Acids to Protocols Amenable to High-Throughput
Screening
Diana Leung and Eric V. Anslyn*
Department of Chemistry and Biochemistry, The UniVersity of Texas at Austin, 1 UniVersity
Station A5300, Austin, Texas 78712
Received May 21, 2008; E-mail: anslyn@ccwf.cc.utexas.edu
Abstract: Enantioselective indicator displacement assays (eIDAs) for R-amino acids were conducted in a
96-well plate format to demonstrate the viability of the technique for the high-throughput screening (HTS)
of enantiomeric excess (ee) values. Chiral receptors [CuII(1)]2+ and [CuII(2)]2+ with the indicator chrome
azurol S were implemented for the eIDAs. Enantiomeric excess calibration curves were made using both
receptors and then used to analyze true test samples. These results were compared to those previously
obtained with a conventional UV-vis spectrophotometer, and they showed little to no loss of accuracy,
while the speed of analysis was increased. A sample of valine of unknown ee was synthesized through an
asymmetric reaction to produce a realistic reaction sample, which was analyzed using receptor [CuII(1)]2+
.
The experimentally determined ee using our eIDA was compared to that obtained by chiral HPLC and 1H
NMR chiral shift reagent analysis. This gave errors of 4.7% and 12.0%, respectively. In addition to the use
of ee calibration curves, an artificial neural network (ANN) was used to determine the % L-amino acid of
the test samples and of the sample of valine of unknown ee from the asymmetric reaction. This method
obtained errors of 5.9% and 2.2% compared to chiral HPLC and 1H NMR chiral shift reagent analysis,
respectively. The technique using calibration curves for the determination of ee on a 96-well plate allows
one to determine 96 ee values in under a minute, enabling its use for HTS of asymmetric reactions with
acceptable accuracy.
Introduction
parallel synthesis and subsequent analysis of a large number of
chiral catalysts. A severe limitation in this method is the large
While asymmetric catalysis is well-recognized as a cost-
effective method for the production of enantiomerically pure
products, one current limitation in catalyst discovery is the
determination of the enantiomeric excess (ee) of reaction
products in a high-throughput (HT) fashion.1-5 Reactions that
yield enantiomerically pure products save money in the phar-
maceutical industry because purification to obtain the desired/
effective enantiomer is avoided and less waste is generated
through discarding of its enantiomer. Asymmetric catalysis can
overcome these disadvantages through direct production of
enantiomerically pure products. Traditional methods to discover
asymmetric catalysts are based on rational design, knowledge
of the reaction mechanism, molecular modeling, and optimiza-
tion through trial-and-error to produce the desired activity.1
In the 1990s, combinatorial libraries came into use to screen
for efficient catalysts.6-9 Combinatorial libraries generally use
number of samples that require screening to determine the best
catalyst.1,10 The use of combinatorial libraries to obtain effective
asymmetric catalysts has not yet been fully utilized due to the
lack of truly high-throughput screening (HTS) methods for the
determination of ee.10 The commonly used methods to determine
ee in a HT fashion are chiral high-performance liquid chroma-
tography (HPLC) and gas chromatography (GC), or HPLC
coupled with circular dichroism (CD).1,11,12 The instruments
used are expensive and only handle a limited number of samples
per day.3 A fairly rapid chiral HPLC run could take 10 min,
though it is often longer, which theoretically translates into
analysis of roughly 144 samples with a HPLC running 24 h a
day.3,10 Also, this does not account for requilibration of the
column between analysis, blank runs, and refilling solvents. True
HTS should accommodate thousands of samples a day. A
number of other methods developed for the determination of
ee have been reviewed recently.1,2,12 Some of these methods
have been transitioned to the microwell plate format, allowing
for rapid screening of ee; however, most need derivatization of
the analytes, which is not compatible with a HTS method.
(1) Reetz, M. T. Angew. Chem., Int. Ed. 2001, 40, 284–310.
(2) Tsukamoto, M.; Kagan, H. B. AdV. Synth. Catal. 2002, 344, 453–
463.
(3) Traverse, J. F.; Snapper, M. L. Drug DiscoVery Today 2002, 7, 1002–
1012.
(4) Stambuli, J. P.; Hartwig, J. F. Curr. Opin. Chem. Biol. 2003, 7, 420–
426.
(9) Gennari, C.; Piarulli, U. Chem. ReV. 2003, 103, 3071–3100.
(10) Kuntz, K. W.; Snapper, M. L.; Hoveyda, A. H. Curr. Opin. Chem.
Biol. 1999, 3, 313–319.
(5) Collins, A. N.; Sheldrake, G. N. C. J. Chirality in Industry; Wiley:
New York/Chichester England, 1992.
(6) Van Arnum, P. Pharmaceutical Technol. 2006, 30, 42–45.
(7) Jaekel, C.; Paciello, R. Chem. ReV. 2006, 106, 2912–2942.
(8) Ding, K. Pure Appl. Chem. 2005, 77, 1251–1257.
(11) Charbonneau, V.; Ogilvie, W. W. Mini-ReV. Org. Chem. 2005, 2, 313–
332.
(12) Finn, M. G. Chirality 2002, 14, 534–540.
9
12328 J. AM. CHEM. SOC. 2008, 130, 12328–12333
10.1021/ja8038079 CCC: $40.75
2008 American Chemical Society

High-Throughput Screening of Enantiomeric Excess Values
A R T I C L E S
an acceptable accuracy for preliminary determination of ee. In
this study, we demonstrate eIDA’s ability as a HTS method by
conducting these studies on a microwell plate, enabling a more
rapid and simpler method of analysis. To transition from a
conventional spectrophotometer to a 96-well plate, ee calibration
curves were regenerated, and analyses of independent test
samples were conducted. The transition to the microwell plate
format from the conventional UV-vis spectrophotometer
showed no significant loss of accuracy upon analysis of test
samples, opening a path to HTS.
Also, to demonstrate eIDAs’ practicability, a sample from
an asymmetric reaction was analyzed with the developed system.
A sample of valine was synthesized through an asymmetric
reaction, and its ee was determined using the chiral receptor
[CuII((R,R)-1)]2+ (Figure 1). This value was compared to those
obtained by chiral HPLC and 1H NMR chiral shift reagent
analysis,23 showing that an eIDA system can compare favorably
to the standard methods and yet function in a HT fashion.
Furthermore, an advanced chemometric tool, an artificial
neural network (ANN), was explored for the determination of
% L-amino acid. An ANN is a data analysis model capable of
modeling complex relationships between the input and output
data. The network’s adaptive system adjusts its structure on the
basis of established learning sets (e.g., calibration data) and can
then be used to apply the learned knowledge to new situations
(unknown samples).24 The use of an ANN to determine %
L-amino acid of test samples and the asymmetrically synthesized
sample of valine allows for an alternative method for determi-
nation of ee.
Figure 1. Structures of ligand 1, chiral receptors [CuII((R,R)-1)]2+ and
[CuII((S,S)-1)]2+, ligand 2, chiral receptors [CuII((R,R)-2)]2+ and [CuII((S,S)-
2)]2+, and indicator chrome azurol S (CAS).
To overcome these disadvantages, the use of a colorimetric
method, such as enantioselective indicator displacement assays
(eIDAs), has been implemented for determining ee. Indicator
displacement assays (IDAs)13,14 have been widely explored by
the chemical community and have also been used as a method
to determine ee.14-20 The use of eIDAs has many advantages
over other methods.21 Because it is based on a colorimetric
measurement of displaced indicators, it uses a conventional
UV-vis spectrophotometer. A spectrophotometer is a relatively
standard instrument in most laboratories, it requires less training,
and it can read samples in a shorter period of time compared to
a chiral HPLC, allowing for screening of asymmetric catalysts
in a HT fashion. Another advantage of using a spectrophotom-
eter is the ability to transition the assay to a microwell plate
reader, allowing for HTS. In a 96-well plate reader, a single
wavelength absorption reading can be obtained in 1 min for
the entire plate. A higher-density plate could also be used,
possibly allowing for a further increase in the number of samples
screened in a given amount of time.
Results and Discussion
1. Design Criteria. After informal consultation with various
individuals in pharmaceutical firms, we have come to realize
that any catalytic system yielding a true ee of 90% or higher in
the screening process should be considered as a promising lead
and selected for analysis with a more accurate technique, such
as chiral HPLC. With that in mind, we have set for ourselves
an upper limit of around 15% absolute error as an accuracy
guideline. We feel that a significantly higher error would render
our method less useful for HTS. Taking this 15% arbitrary limit
into consideration, any sample with an ee below 75% of the
desired enantiomer could be discarded outright, because the ee
values above 75% could possibly fall near true ee values of
90%. Given that we define ee to span -100% to 100%, this
means that if the ee values of unknowns were evenly distributed,
our method would allow one to discard 75% of the samples
(Figure 2a) and submit only the best leads to time-consuming
HPLC analysis. The high negative values should also be
analyzed because the enantiomeric catalyst could be used.
Further, in a case where the ee values obtained from random
catalysts screening were normally distributed (Gaussian), an
even lower fraction of samples would be retained for analysis
(Figure 2b). This would considerably decrease the overall
screening time without compromising final accuracy. Of course,
lower errors would allow for an even higher ability to narrow
the selection of samples for analysis by a more accurate method,
and slightly higher errors would lead to more samples to be
analyzed. We interpret the results given herein having this goal
in mind.
In our previous paper, we have shown that the use of eIDAs
with two receptors ([CuII(1)]2+ and [CuII(2)]2+) and an indicator,
chrome azurol S (CAS, see Figure 1), can enantioselectively
discriminate 13 of the 17 analyzed R-amino acids.22 Enantio-
meric excess of true test samples could be determined on a
UV-vis spectrophotometer in aqueous media at pH 7.5, with
(13) Lavigne, J. J.; Anslyn, E. V. Angew. Chem., Int. Ed. 1999, 38, 3666–
3669.
(14) Nguyen, B. T.; Anslyn, E. V. Coord. Chem. ReV. 2006, 250, 3118–
3127.
(15) Folmer-Andersen, J. F.; Kitamura, M.; Anslyn, E. V. J. Am. Chem.
Soc. 2006, 128, 5652–5653.
(16) Folmer-Andersen, J. F.; Lynch, V. M.; Anslyn, E. V. J. Am. Chem.
Soc. 2005, 127, 7986–7987.
(17) Zhu, L.; Anslyn, E. V. J. Am. Chem. Soc. 2004, 126, 3676–3677.
(18) Zhu, L.; Shabbir, S. H.; Anslyn, E. V. Chem.-Eur. J. 2006, 13, 99–
104.
(19) Kacprzak, K.; Grajewski, J.; Gawronski, J. Tetrahedron: Asymmetry
2006, 17, 1332–1336.
(20) Kubo, Y.; Ishida, T.; Kobayashi, A.; James, T. D. J. Mater. Chem.
2005, 15, 2889–2895.
(21) Zhu, L.; Zhong, Z.; Anslyn, E. V. J. Am. Chem. Soc. 2005, 127, 4260–
4269.
(23) Inamoto, A.; Ogasawara, K.; Omata, K.; Kabuto, K.; Sasaki, Y. Org.
Lett. 2000, 2, 3543–3545.
(22) Leung, D.; Folmer-Andersen, J. F.; Lynch, V. M.; Anslyn, E. V. J. Am.
Chem. Soc. 2008, 130, 12318-12327 (preceding paper in this issue).
(24) Peterson, K. L. ReV. Computa. Chem. 2000, 16, 53–140.
9
J. AM. CHEM. SOC. VOL. 130, NO. 37, 2008 12329

A R T I C L E S
Leung and Anslyn
Figure 2. (a) Supposing that the ee products from an asymmetric reaction
were evenly distributed, with a (15% error in ee values, approximately
75% could be discarded after screening. (b) In the case of a Gaussian
distribution of product ee, an even lower number of samples would need
to be further analyzed with a slower, high-accuracy technique.
2. Determining the Enantiomeric Excess of Test Samples
on a 96-Well Plate Reader. Chiral ligand 1 or 2 coordinates to
a CuII metal center, leaving two open coordination sites for the
exchange between the indicator CAS and R-amino acids.16 In
the preceding paper, we described the ability to produce ee
calibration curves and to determine ee values of test samples
with a standard UV-vis spectrophotometer by exploiting the
different stabilities of diastereomeric complexes created by
[CuII((R,R)-1)]2+ and [CuII((R,R)-2)]2+ with chiral amino ac-
ids.22 The use of a 96-well plate was examined here as a means
to integrate a rapid and automated method for determining ee,
which will allow for a truly HTS of asymmetric catalysts. All
ee calibration curves and test samples were prepared with a
BioTek Precision Microplate Pipetting System on 96-well plates
(Figure 3).
The same amino acids analyzed with a UV-vis spectropho-
tometer in the preceding paper for these two receptors were
also analyzed here.22 Because the 96-well plate methodology
is less labor-intensive and faster, we decided to add additional
amino acids for analysis. In order to broaden the scope of our
method, aspartate was added to analyses conducted with
[CuII((R,R)-1)]2+ and tryptophan for [CuII((R,R)-2)]2+. On a 96-
well plate, four rows were used for ee calibration curves, one
for each amino acid.25,26 In the remaining four rows, six test
samples for each amino acid were laid in each row, one for
each amino acid. As representative examples, two ee calibration
curves are shown in Figure 4 for each receptor. The test samples’
ee values were determined using the corresponding calibration
curve. These test samples were made independently of the ee
calibration curve, therefore affording a more convincing valida-
tion rather than the use of a jack-knife analysis.
Figure 3. The 96-well plate used for making the ee calibration curves. (a)
For valine, aspartate, histidine, and isoleucine using receptor [CuII((R,R)-
1)]2+ (top four rows) and six test samples for each of the above amino
acids (bottom four rows). (b) For valine, tryptophan, alanine, and serine
using receptor [CuII((R,R)-2)]2+ (top four rows) and six test samples for
each of the above amino acids (bottom four rows).
for analyses conducted with [CuII((R,R)-1)]2+ and [CuII((R,R)-
2)]2+ (Table 1). The overall average of these individual average
absolute errors was calculated to be (9.7% for receptor
[CuII((R,R)-1)]2+ and (26.6% for receptor [CuII((R,R)-2)]2+
,
which is comparable to or somewhat higher than what was
obtained with a standard UV-vis spectrophotometer analysis
((10.2% and (13.6%, respectively) in our previous study (see
preceding paper).22 Unfortunately, most of the average absolute
errors using [CuII((R,R)-2)]2+ exceeded our arbitrary limit of
15% and would require improvements for such a system to be
applied to them, but valine had an acceptable error of 8.5%.
On the other hand, the average absolute errors using [CuII((R,R)-
1)]2+ were well within the limit, as is exemplified by the case
of histidine (4.8%). The transition from the UV-vis spectro-
photometer to a 96-well plate reader has allowed for a much
faster and automated process of analysis, since it takes ap-
proximately 1 min to read an entire plate, although the error
suffered when [CuII((R,R)-2)]2+ was used.
The transition from the UV-vis spectrophotometer to a
microwell plate format showed little loss of accuracy when
using [CuII((R,R)-1)]2+ but a loss of accuracy with the use of
[CuII((R,R)-2)]2+. The observed increase in error when using
[CuII((R,R)-2)]2+ could be due to the decrease in ∆Amax values
when transitioning to the microwell plate format from a single-
cell cuvette. Upon transitioning, a decrease in the path length
occurred, which would decrease the absorbance, leading to the
compression of the range of absorbances between -100% and
100% ee. As discussed in the previous paper, the errors usually
correlate to the ∆Amax values: the larger the ∆Amax values, the
larger the difference in absorbance between two mixtures with
close ee, thus lowering the errors.22 Since the ∆Amax values
obtained for [CuII((R,R)-2)]2+ on the UV-vis spectrophotometer
The ee calibration curves were subjected to linear and second-
order polynomial regression. The best-fit curve (refer to Sup-
porting Information) was used to determine the ee’s of true test
and unknown samples. The average absolute errors were
computed by taking the absolute difference between the actual
and the experimental ee values and averaging them for the six
test samples. The average of the absolute errors was obtained
(25) Refer to Supporting Information, Figure S-1, for ee calibration curves
of a selection of amino acids analyzed with [CuII((R,R)-1)]2+ on a
96-well plate reader.
(26) Refer to Supporting Information, Figure S-2, for ee calibration curves
of a selection of amino acids analyzed with [CuII((R,R)-2)]2+ on a
96-well plate reader.
9
12330 J. AM. CHEM. SOC. VOL. 130, NO. 37, 2008

High-Throughput Screening of Enantiomeric Excess Values
A R T I C L E S
Table 1. Average Absolute Errors for the Determination of ee of
Test Samples on a 96-Well Plate Reader Using Two Receptors
(a) [CuII((R,R)-1)]2+a
(b) [CuII((R,R)-2)]2+b
Amino Acid
average absolute error (%)
Amino Acid
average absolute error (%)
Asp
His
Ile
13.5
4.8
10.3
10.2
Ala
Ser
Trp
Val
32.0
47.6
18.3
8.5
Val
a Average absolute error of ee determination of six test samples for
each amino acid analyzed through absorbance measurements taken on a
96-well plate reader, and ee calibration curves made for each amino acid
using receptor [CuII((R,R)-1)]2+. Test samples containing CAS (10 µM),
Cu(OTf)2 (200 µM), (R,R)-1 (2.5 mM), and a mixture of L- and D-amino
acids (different concentrations of amino acids were used, refer to Table
S-1 in the Supporting Information) were mixed and diluted to 300 µL
with 1:1 MeOH:H2O, buffered to pH 7.5 with 50 mM HEPES. For
experimental values obtained, refer to Supporting Information Table S-3.
b Average absolute error of ee determination of six test samples for each
amino acid analyzed through absorbance measurements taken on a
96-well plate reader, and ee calibration curves made for each amino acid
using receptor [CuII((R,R)-2)]2+. Test samples containing CAS (10 µM),
Cu(OTf)2 (105 µM), (R,R)-2 (8.8 mM), and a mixture of L- and D-amino
acids (different concentrations of amino acids were used, refer to Table
S-2 in the Supporting Information) were mixed and diluted to 300 µL
with 1:1 MeOH:H2O, buffered to pH 7.5 with 50 mM HEPES. For
experimental values obtained, refer to Supporting Information Table S-4.
were already close to 0.1, which was our set threshold for
enantioselectivity, some of them decreased below 0.1 when
transitioning to a microwell plate format due to the decrease in
path length. With the ∆Amax values below 0.1, an increase in
errors was observed. For receptor [CuII((R,R)-1)]2+, the ∆Amax
values obtained in the UV-vis spectrophotometer were well
over 0.1, so that the decrease in path length did not cause them
to drop below 0.1, thus not affecting the errors. Actually, the
errors decrease for receptor [CuII((R,R)-1)]2+ due to the use of
automated liquid handlers, which allowed for an increase in
reproducibility. It is important to note that the path length issue
could be corrected with the use of a deeper plate, which is
commercially available. Also, the transition to a higher density
plate does not require a reduction in path length.
3. Analysis of Asymmetrically Synthesized Unknown ee
Sample. To show the practicability of the proposed method, the
eIDA was used for the analysis of samples obtained from an
asymmetric reaction. In order to provide such a sample, valine
was synthesized using the asymmetric reaction shown in Scheme
1.27,28 The ee value of the unknown valine sample was first
determined using the eIDA method with [CuII((R,R)-1)]2+, and
this value was compared to that obtained by two well-accepted
1
protocols: HPLC using a chiral stationary phase and H NMR
with the use of a chiral shift reagent.23
On a 96-well plate, an ee calibration curve for valine using
[CuII((R,R)-1)]2+ was laid out on one row, with another row
containing six replicates of a sample of valine of unknown ee
obtained from the reaction. The ee’s of the six samples were
computed using the best-fit curve of the ee calibration curve
and averaged to be 84.6% (σ ) 4.9%). The unknown valine
sample was then analyzed by chiral HPLC using the chiral
column Daicel Crownpak CR(+) (refer to the Supporting
Information for analysis conditions) to give an ee of 79.9%.
The Daicel Crownpak CR(+) was used in the analysis because
it is the chiral HPLC column most commonly used for R-amino
Figure 4. Enantiomeric excess calibration curves obtained using a 96-
well plate reader. (a,b) Absorbance at 602 nm as a function of ee for
displacement experiments performed in a solution containing CAS (10 µM),
Cu(OTf)2 (200 µM), and (R,R)-1 (2.5 mM) in 1:1 MeOH:H2O, 50 mM
HEPES buffered to pH 7.5, with the addition of (a) aspartate (302 µM) or
(b) histidine (202 µM). (c,d) Absorbance at 602 nm as a function of ee for
displacement experiments performed with the addition of amino acid into
a solution containing CAS (10 µM), Cu(OTf)2 (105 µM), and (R,R)-2 (8.8
mM) in 1:1 MeOH:H2O, 50 mM HEPES buffered to pH 7.5, with the
addition of (c) alanine (149 µM) or (d) valine (125 µM).
(27) Chakraborty, T. K.; Hussain, K. A.; Reddy, G. V. Tetrahedron 1995,
51, 9179–9190.
(28) Inaba, T.; Kozono, I.; Fujita, M.; Ogura, K. Bull. Chem. Soc. Jpn.
1992, 65, 2359–2365.
9
J. AM. CHEM. SOC. VOL. 130, NO. 37, 2008 12331

A R T I C L E S
Leung and Anslyn
Scheme 1. Asymmetric Reaction Used To Produce a Sample of
Valine of Unknown ee27,28
Table 3. Average Absolute Errors for the Determination of %
L-Amino Acid and ee, Using ANN with Data Collected on a
96-Well Plate Reader Using Two Receptors
(a) [CuII((R,R)-1)]2+
(b) [CuII((R,R)-2)]2+
absolute error (%)
absolute error (%)
Amino Acid
for % L-AA
for ee
Amino Acid
for % L-AA
for ee
Asp
His
Ile
4.4
5.0
6.6
4.0
8.8
9.9
13.1
8.0
Ala
Ser
Trp
Val
16.4
20.3
27.8
3.8
32.9
40.6
55.6
7.6
Val
Scheme 2. Equations Used To Determine Enantiomeric Excess
Using % L-Amino Acid Determined through ANN Analysis
[L] - [D]
ee )
(1)
(2)
(3)
[L] + [D]
(%L) - (100 - (%L))
ee )
100
2(%L ( error) - 100
ee )
acids. Compared to chiral HPLC, the results of the eIDA method
were shown to differ by 4.7%, which is within one standard
deviation of the eIDA value. The ee of the unknown valine
sample was also determined by an NMR titration, using the
ratio of the integral of the HR’s of L- and D-valine when mixed
with sodium[(R)-1,2-diaminopropane-N,N,N′,N′-tetraacetato]sa-
marate(III) hydrate23 (8, Figure 5). It was determined to be
72.6%, which is 7.3% off from the HPLC result. The eIDA
analysis thus differed by 12.0% from the NMR results. This
comparison affirmed that a sample from an asymmetric reaction
could be analyzed with the proposed eIDA with acceptable
accuracy and an error well within the range we have set for
ourselves for preliminary screening of catalysts in asymmetric
reactions in HT fashion with the use of a 96-well plate reader
(Table 2).
4. Artificial Neural Network Analysis of Data. Our group has
previously shown that an ANN can be used to lower the absolute
errors of eIDAs.18 Therefore, an ANN was used as a second
method of analysis of the amino acid test samples. Statistica
ANN software, version 4.0 E, was used to train a different
network for each amino acid with the data collected for ee
calibration curves on a 96-well plate reader. A different network
was required for each amino acid because each one has a
different ee calibration curve.
100
training sets (absorbance between 570 and 620 nm). The range
of wavelengths used to train the network was restricted to that
containing the relevant spectral changes (570-620 nm). The
ANN system was trained on the basis of the percentage of one
enantiomer rather than ee, simply because the software is unable
to accommodate negative values, so % L-amino acid was used.
Percentage of L-amino acid is defined as the concentration of
L-amino acid, divided by the total concentration of L- and
D-amino acids.
Networks were developed for each amino acid, and the %
L-amino acid values of the test samples analyzed with [CuII-
((R,R)-1)]2+ and [CuII((R,R)-2)]2+ were determined by inputting
the test samples absorbances at the wavelengths with which the
network was trained (570-620 nm), in order to check the
efficacy of the ANN approach. Since the ee calibration curves
are different for each amino acid and two receptors were used
in the analysis, a different network is required for each one.
An average absolute overall error of (5.0% was obtained using
receptor [CuII((R,R)-1)]2+ and (17.1% with [CuII((R,R)-2)]2+
which are acceptable errors. The average absolute error for each
amino acid for determining % L-amino acid with an ANN is
shown in Table 3.
,
To train a network, the raw data already collected for the
purpose of establishing ee calibration curves were used as
The errors obtained for the test samples using ee calibration
curves may appear to be higher than the errors obtained using
ANN, but these two methods cannot be compared directly since
they are determining two different variables (ee and % L-amino
acid). However, given the way we calculated ee (Scheme 2, eq
1), it is possible to compute ee using the % L-amino acid
determined with ANN analysis (Scheme 2, eq 2). Numerically,
the errors for ee will appear to double due to the very definition
of this parameter (Scheme 2, eq 3). This is demonstrated by
converting the % L-amino acid determined using ANN to ee
(Scheme 2, eq 2) and by recalculating their average absolute
errors, which are also shown in Table 3. The error in ee is double
the error obtained with analysis using ANN for determining %
L-amino acid (Table 3). After determining the ee using the value
of % L-amino acid determined through ANN, the overall average
absolute errors are (10.0% and (34.2% for analyses conducted
with [CuII((R,R)-1)]2+ and [CuII((R,R)-2)]2+, respectively. This
is comparable to the errors obtained for determining ee using
ee calibration curves (Table 1).
Figure 5. Chiral shift reagent, sodium[(R)-1,2-diaminopropane-N,N,N′,N′-
tetraacetato]samarate(III) hydrate (8).
Table 2. Determination of ee of a Sample of Valine of Unknown
ee Using eIDA ([CuII((R,R)-1)]2+), Chiral HPLC, and 1H NMR
Chiral Shift Reagent
ee (%)
by eIDAa
by Chiral HPLCb
by Chiral Shift Reagentc
unknown sample
84.6
79.9
72.6
a Average ee determined from six replicates of a sample of valine of
unknown ee using receptor [CuII((R,R)-1)]2+ on a 96-well plate reader.
b ee determined through HPLC analysis with chiral column, Daicel
Crownpak CR(+). c ee determined through 1H NMR analysis with a
chiral shift reagent, 8.
The raw data previously obtained in a 96-well plate format
for the six replicate valine samples of unknown ee, synthesized
9
12332 J. AM. CHEM. SOC. VOL. 130, NO. 37, 2008

High-Throughput Screening of Enantiomeric Excess Values
A R T I C L E S
Table 4. ANN Analysis To Determine % L-Valine of a Sample of
Valine of Unknown ee on a 96-Well Plate Reader
reactions that generated low-ee products could be discarded,
while only the best leads (for example, those whose products
had 90+% ee within the error of our analysis) would be further
examined, and their products would be reassessed with a more
accurate standard technique to be taken on to the next level of
optimization. Such a two-tier method would expedite the
screening process immensely.
ANNa
chiral HPLC
chiral shift reagent
% L-unknown valine
84.1
90.0
86.3
a Network trained with raw data obtained from the ee calibration
curve for valine, registered on a 96-well plate reader using receptor
[CuII((R,R)-1)]2+
.
Summary
through an asymmetric reaction, were also analyzed using the
appropriate ANN to determine the % L-amino acid. These
samples averaged to be 84.1% (σ ) 1.0%) L-valine, which
corresponds to an ee of 68.2%. The errors between the %
L-valine value calculated by ANN and those obtained with the
chiral HPLC and chiral shift reagent23 are 5.9% and 2.2%,
respectively. The comparison of % L-valine determined through
ANN, chiral HPLC, and chiral shift reagent is shown in Table
4. This shows the high reliability of our approach in determining
unknown % L-amino acid when the chiral receptor gives low
overall errors.
The system presented here is truly a HTS technique because,
when a single-wavelength-based calibration curve is employed,
the time required for measuring the absorbance at one wave-
length of all 96-wells is approximately 1 min. On a 96-well
plate, we generally laid out four calibration curves, but in a
typical HTS, there is one specific analyte of interest, so only
one row would be taken up by the calibration curve, whereas
the rest of the microwell plate would be used for samples. The
calibration curves thus obtained could be automatically used to
determine the ee of the samples, which requires less than a
minute with a computer. This means that the analysis time of
84 samples would require 2 min, taking into account the
measurement and processing of data. Liquid handlers can be
used to dispense solutions into 96-well plates, as was done in
this study. Also, an automated plate stacker can remove and
place a new plate into the reader, thereby completely automating
the procedure: this would allow for analysis of samples from
start to finish without any operator intervention.
In cases where ANN was used as the method of analysis, a
range of wavelengths would be scanned: it takes approximately
9 min to obtain absorbance data from 570 to 620 nm for the
entire 96-well plate. The absorbance at this range of wavelengths
could be used automatically by a computer to generate and train
the ANN. It takes approximately 15 min to create a neural
network that best models the training set. Hence, for the analysis
of one amino acid (i.e., one calibration curve and seven rows
of samples), it would require approximately 25 min to obtain
measurements of the entire plate (absorbance ) 570-620 nm)
and to train to obtain the best neural network. The computer
could then compute the % L-amino acid with a trained network
in hand. Once a network is generated, it does not need to be
recomputed as long as instrumental parameters do not change,
and the analysis time will be shortened to the time required to
read 96-wells and process the read data, which is around 10
min.
In this study, enantioselective indicator displacement assays
(eIDAs) have been used to determine the enantiomeric excess
of R-amino acids with acceptable accuracy using a 96-well plate
reader, as would be done in high-throughput screening. Enan-
tiomeric excess was determined for test samples using ee
calibration curves on a 96-well plate reader, affording an overall
average error of (9.7% when analysis of samples was conducted
with receptor [CuII((R,R)-1)]2+ and an overall average error of
(26.6% when receptor [CuII((R,R)-2)]2+ was used. As shown
in this paper, the errors produced through the analysis of test
samples on a 96-well plate reader were consistent with those
obtained on a UV-vis spectrophotometer, with the advantage
of increased speed of analysis, as required in a HTS method.
A sample of valine was asymmetrically synthesized, and its
ee, analyzed with receptor [CuII((R,R)-1)]2+, was determined
to be 84.6%. This result was compared to that obtained by
1
reference methods (chiral-column HPLC and H NMR chiral
shift reagent) and found to be in good agreement (79.9% and
72.6%, respectively). This showed the assay’s ability to analyze
real-world samples.
Artificial neural networks were used as a second method to
determine the % L-amino acid of test samples. Using the
collected absorbance data, networks were developed for each
amino acid. ANN analysis on the data collected for test samples
in a 96-well plate format showed overall average errors of
(5.0% and (17.1% for analyses conducted with [CuII((R,R)-
1)]2+ and [CuII((R,R)-2)]2+, respectively. ANN was also able
to determine % L-amino acid of the sample of valine of unknown
ee with an excellent error relative to standard methods, chiral
HPLC and chiral shift reagent, giving 5.9% and 2.2% errors,
respectively.
In conclusion, the proposed eIDA method has been shown
to be capable of determining ee of R-amino acids as a HTS
technique mainly due to the use of colorimetric sensing. The
transition to a microplate reader from a UV-vis spectro-
photometer was possible, allowing for a more rapid analysis
time while retaining accuracy in determining ee, showing that
the method is robust. Even a real-world sample from a reaction
showed excellent agreement with accepted methods. The use
of a microwell plate reader allows for a seamless integration
into a HTS workflow aimed at asymmetric reaction discovery
or optimization.
Acknowledgment. We gratefully acknowledge financial support
from the National Institutes of Health (GM77437) and the Welch
Foundation.
The method proposed here is much more rapid than the use
of chiral HPLC. As discussed in the Design Criteria section,
the purpose that we envision does not strictly require extreme
accuracy. In fact, in the first steps of a screening for asymmetric
catalysts, even a pass/fail answer based on a threshold ee value
would be adequate. After screening with our method, the
Supporting Information Available: Experimental details,
calibration curves, ee determinations, and artifical neural
network analysis. This information is available free of charge
via the Internet at http://pubs.acs.org.
JA8038079
9
J. AM. CHEM. SOC. VOL. 130, NO. 37, 2008 12333