Use of molecular recognition to drive chemical evolution: Mechanisms of an automated genetic algorithm implementation

Article abstract of DOI:10.1002/(SICI)1521-3765(19980515)4:5<825::AID-CHEM825>3.0.CO;2-7

A detailed description is provided for a new general approach that allows production of amplified amounts of effective noncovalent binders from pools of compounds that can exist in a dynamic equilibrium. This approach involves a) selective binding of the

Full text of DOI:10.1002/(SICI)1521-3765(19980515)4:5<825::AID-CHEM825>3.0.CO;2-7

FULL PAPER
Use of Molecular Recognition To Drive Chemical Evolution:
Mechanisms of an Automated Genetic Algorithm Implementation**
Alexey V. Eliseev* and Marina I. Nelen
Dedicated to Professor Fredric M. Menger on the occasion of his 60th birthday
Abstract: A detailed description is pro-
vided for a new general approach that
allows production of amplified amounts
of effective noncovalent binders from
pools of compounds that can exist in a
dynamic equilibrium. This approach in-
volves a) selective binding of the effec-
tive pool components by the immobi-
lized target compound; b) equilibration
of the pool remaining after selection,
and thereby regeneration of the effec-
tive components; c) repetition of the
selection ± equilibration cycles. It repre-
sents a new, automated, implementation
of a genetic algorithm/mechanism in a
chemical system. Application of the
approach to the generation of arginine
binders involved the synthesis and
screening of eight compounds, each of
which was capable of forming a pool of
three isomers (trans,trans, trans,cis, and
cis,cis) that could be interconverted by
photoisomerization. These compounds
were screened for their affinity for, or
selectivity of binding to, guanidinium
derivatives. The most promising com-
pound was then used in our method to
generate an amplified amount of the
cis,cis isomer, the strongest binder from
the equilibrating pool. The receptors
were selected using an arginine deriva-
tive immobilized on the modified silica
gel support, which had been found to
possess binding affinity and selectivity
similar to those of guanidinium salts in
solution. Mathematical models of the
approach were developed that allowed
us to evaluate the importance of various
experimental parameters and to assess
the applicability of the method to larger
combinatorial pools.
Keywords: combinatorial chemistry
´
genetic algorithms ´ molecular
recognition ´ photochemistry ´ tem-
plate synthesis
The general sequence of events used by GAs is shown in
Figure 1. The initial population, which is simply represented in
a chemical system by an array of compounds, undergoes
Introduction
Among the numerous lessons chemists have learned from
biology, that of evolutionary development is probably still the
most difficult to follow. Applications of the principles of
Darwinian evolution to a wide variety of scientific problems
were formalized long ago in terms of so-called genetic
algorithms (GAs) which, until recently, have been mostly
applied to solving problems in the field of computer science.[1]
During the last decade, however, new techniques emerging in
chemistry and biochemistry, as well as a vastly increased
diversity of molecular structures dealt with on a routine basis,
have drawn chemistsꢀ attention to the use of GAs as a method
that helps to generate or find novel compounds with the
required properties.[2]
Figure 1. General scheme of the genetic algorithms.
[*] Prof. Dr. A. V. Eliseev, Dr. M. I. Nelen
Department of Medicinal Chemistry, School of Pharmacy
State University of New York at Buffalo
Buffalo, NY 14260 (USA)
multiple cycles of selection ± reproduction to generate a new
array (or subarray), possessing new (or improved) properties,
such as binding affinity for a particular target compound or
the ability to catalyze a chemical reaction.
Fax: (1)716-645-2393
E-mail: eliseev@acsu.buffalo.edu
There are several major approaches aimed at the imple-
mentation of various attributes of the GA in chemical and
[**] Use of Molecular Recognition to Drive Chemical Evolution, Part 2.
For Part 1, see ref. 9.
Chem. Eur. J. 1998, 4, No. 5
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0825 $ 17.50+.25/0
825

A. V. Eliseev and M. I. Nelen
FULL PAPER
biochemical systems. The most remarkable among them
include recent work on the optimization of combinatorial
libraries,[3] evolutionary enrichment of large nucleic acid
pools,[4] mutagenesis/in vivo selection of proteins,[5] and the
chemistry of self-replicating systems.[6]
Figure 2) than the majority of the pool components. If the
target compound is then introduced into the equilibrating
pool, the equilibrium should be shifted towards the formation
of complexes Ai ´ T corresponding to the components possess-
ing higher binding constants with the target.[11] In the ideal
situation shown in Figure 2, the strong complexation of one of
the components (Ai) will drive the system to the production of
an amplified amount of Ai, in the form of the complex Ai ´ T, at
the expense of the other components in the pool.
Whereas a targeted equilibrium shift of this kind is
essentially predicated on the well-known Le Chatelier prin-
ciple, its chemical realization may open new routes to
generation of combinatorial libraries, leading to discoveries
of novel synthetic ligands and receptors. Thus, Venton
et al.[12a] and Swann et al.[12b] have introduced a method for
using macromolecules as molecular traps for components
selected from reversibly synthesized peptide libraries, and
have detected amplification of the higher affinity ligands. The
utility of such an approach has recently been demonstrated by
Huc and Lehn[13] in the formation of virtual libraries of
functional imines that have been enriched with carbonic
anhydrase inhibitors upon introduction of the target enzyme.
Sanders et al. have reported examples of the thermodynami-
cally controlled formation of select macrocycles containing
complementary building blocks, as opposed to a multitude of
theoretically possible compounds.[14]
Some of the problems that may arise in practical applica-
tions of the targeted equilibrium shift in equilibrating pools
are related to the presence of the target compound in the
mixture. Indeed, one can envision a variety of side-interac-
tions, of covalent or noncovalent types, that may interfere
with the binding and equilibration processes. In addition, the
target compounds, particularly biopolymers, may be sensitive
to the equilibration conditions. Finally, the free and bound
pool components always have to be separated from the target.
The experimental method introduced here to overcome the
problems mentioned above involves a physical separation of
the equilibration and binding sites. In the reaction setup
shown in Figure 3, the pool is circulated through two
Application of the GA to combinatorial chemistry has been
based on manual implementation, that is, stepwise identifica-
tion and optimization of the effective leads. The other
approaches involve chemical or biological selection and/or
amplification processes. Thus, the selection of structures can
be driven by the thermodynamics of binding (as in nucleic
acid evolution[4]), the kinetics of chemical transformations (as
in self-replication of individual compounds[6], molecular
assemblies,[7] and amplification of chirality in chemical
reactions[8]), or the metabolic processes in cell cultures
(protein evolution[5]). We believe, therefore, that it may be
more appropriate to refer to these approaches as employing
genetic mechanisms, rather than genetic algorithms.
We have recently introduced a novel approach that makes it
possible to amplify a subset of compounds possessing higher
affinity for a given target, within an equilibrating pool of
components.[9] This approach involves a more automated
implementation of the features of GAs, such as selection and
mutation, than the previously known techniques. In this paper
we present an example of a general strategy that can be used
to design an evolving equilibrating pool, describe mechanisms
and mathematical models of this new method, and assess the
applicability of the approach to more complex systems.
Results and Discussion
Principle of the approach: The thermodynamic basis of the
chemical evolution approach is outlined in Figure 2. The most
important requirement for the method is the availability of a
Figure 2. Shift of equilibrium in the dynamic pool by the target compound.
mixture (pool) of components (A1 ± An) (Figure 2) that can be
brought to a dynamic chemical equilibrium by applying
specific conditions. The equilibration can be based on any
kind of reversible chemical reaction, such as photochemical
isomerization (as presented here) or exchange reactions, some
of which are mentioned below. The features of the exper-
imental design of such pools are akin to the general principles
of combinatorial library design;[10] that is, a single component
(Ai, Figure 2) or a small subset of components is expected to
show higher affinity for a particular target compound (T,
Figure 3. Experimental implementation of the dynamic (evolutionary)
molecular diversity approach.
chambers. The selection site (lower chamber) contains the
target compound in its immobilized form, which preferen-
tially binds the effective components of the pool, like an
affinity chromatography column.[15] As a result of passing
826
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0826 $ 17.50+.50/0
Chem. Eur. J. 1998, 4, No. 5

Automated Genetic Algorithms
825±834
through this column the pool is depleted of the effective
components (strong binders). It then enters the upper
chamber, in which the pool components are brought to a
dynamic equilibrium by physical initiation (e.g., by irradiation
with light; see below) or catalysis of the equilibration
reaction. Such an equilibration regenerates the fraction of
the effective component(s) in the pool, which then moves into
the selection chamber again. Repetition of these cycles will
gradually lead to amplification of the effective binder(s) in its
target-cmplexed form and a decreased concentration of the
original mixture in solution. Thus, this method is designed to
reorganize the original pool into a subset of compounds
enriched with the high-affinity binders. As will be shown
below, such stepwise evolutionary-type enrichment leads to
results similar to those that might be expected from one-phase
equilibration of the pool with the target compound, but makes
it possible to avoid undesirable side-reactions between the
target and its binders and to simplify isolation of the high-
affinity subset.
both of these possibilities in our receptors, we synthesized a
series of compounds, 1a ± 1d, 2a, 2b, 3a, and 3b, containing
either anionic groups or hydrogen-bond acceptors capable of
polytopic interactions with the ligand. According to the
molecular design, one of the isomers of each receptor, namely
cis,cis, would have the most favorable geometry for binding to
the ligand because of the closeness of its two arms as defined
by its isomerization motif (Figure 4).
Application of the approach to the generation of arginine
receptors: In developing the approach described, our first
goal was to design a relatively simple system containing a
limited number of components. In such a system, all the
important parameters, such as the affinities of the individual
components for the target, the distribution of the components
in the mixture, their rates of interconversion, could be
separately evaluated and related to the mechanisms of the
evolutionary process. Quantitative assessment of the system
parameters should then make it possible to evaluate the
applicability of the approach to more complex systems.
In order to meet the first requirement of the method,
reversible interconversion of the components, we used
molecular scaffolds 1, 2, and 3, each containing two double
bonds that might undergo cis ± trans isomerization on irradi-
ation with UV light. As defined by the symmetrical structure
of the scaffolds, three isomers, trans,trans, cis,trans, and cis,cis,
would coexist in the irradiated solution.
Figure 4. Proposed binding modes of guanidinium salts to the cis,cis
isomers of compounds 1 ± 3: a) through multiple hydrogen bonds (1b ± 1d,
2b, 3b); b) through the formation of two salt bridges (1a, 2a, 3a).
Receptors 1 ± 3 were synthesized from commercially avail-
able compounds in two to three steps (see Experimental
Section). The double-bond connections between the aromatic
moieties and the functional units were built from the
corresponding carbonyl compounds through the aldol con-
densation ± dehydration reactions (compounds 1c, 1d) and
through the Wadsworth ± Emmons reactions with the phos-
phonate derivatives (compounds 2, 3).
Each potential receptor, initially obtained in its trans,trans
form,[17] was then tested for its applicability to our system. The
following properties were essential: i) the ability to form all
three isomers by reversible light-induced isomerizaton; ii) the
affinity of the cis,cis isomer for the ligand; iii) the selectivity of
binding among the isomers. The isomerization and binding
studies were performed in various media, as dictated by the
solubility of the receptors and the polarity or hydrogen-
bonding properties of the solvents. After irradiation of the
receptor solution with broad-band UV± visible light, the
content of each of the isomers was determined by NMR
spectroscopy and/or HPLC. The receptor ± ligand interactions
were studied and, wherever possible, binding constants were
determined by NMR spectroscopic titrations of the mixture of
isomers with the hydrochloride of the corresponding guani-
dinium derivative. The results of the screening of eight
potential receptor systems are summarized in Table 1.
Besides the apparent biological importance of the guanidi-
nium fragment, the choice of its derivatives as the target
ligands was based on its ability to engage in molecular
recognition in various modes, primarily through the formation
of salt bridges and of hydrogen bonds.[16] In order to explore
The electrostatic binders (1a, 2a, 3a), studied in protic
solvents with methylguanidinium hydrochloride, showed the
highest promise for our approach. Their irradiation led to
comparable amounts of all three isomers, which had appreci-
Chem. Eur. J. 1998, 4, No. 5
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0827 $ 17.50+.25/0
827

A. V. Eliseev and M. I. Nelen
FULL PAPER
Table 1. Results of molecular scaffold screening for the light-isomerizable arginine receptors.
1
Recep- Guest
tor
Solvent
Binding constant (K) [m
]
Selectivity among receptor isomers Method
cis,cis
cis,trans
trans,trans (Kcis,cis/Kcis,trans/Ktrans,trans)
1a
1a
1b
2a
methylguanidinium 10% D2O in [D4]methanol
methylguanidinium [D4]methanol
dodecylguanidinium CDCl3
methylguanidinium [D4]methanol
[D6]ethanol
86
<20
<20
44 Æ 17
n/a
4.9/2.7/1
1H NMR
1H NMR
213 Æ 9 120 Æ 15
[a]
[a]
[a]
170 Æ 12
72 Æ 5
< 3
4 Æ 3
ca. 60/25/1
250/38/1
1H NMR
1H NMR
1H NMR
1H NMR
31P NMR
31P NMR
1H NMR
980 Æ 90 150 Æ 50
10% D2O in [D10]tert-butyl alcohol > 2500
790 Æ 110 40 Æ 25 > 63/20/1
2b
3a
3b
3b
phenylguanidinium CDCl3
methylguanidinium [D4]methanol
phenylguanidinium CDCl3
phenylguanidinium CDCl3
22 Æ 13
7 Æ 1
7 Æ 2
160 Æ 15
35 Æ 17
21 Æ 7
3/1/1
29 Æ 3
40 Æ 5
0.2/0.25/1
0.6/0.6/1
3.5/1.4/1
22 Æ 13 21 Æ 10
73 Æ 26 29 Æ 9
[a] No binding was detected within the guest concentration range 0 ± 0.1m.
able binding affinities for the ligand. The affinities of the
isomers of 1a and 2a were in line with the predictions
obtained from molecular modeling, as described elsewhere.[18]
Surprisingly, the cis,cis and cis,trans isomers of compound 3a
showed comparable affinities for methylguanidinium in
methanol, whereas the trans,trans isomer was a better binder
than the other two. As a possible explanation, we suggest that
the strong electrostatic repulsion between the phosphate
dianions of 3a in the complexation mode shown in Figure 4
makes a highly unfavorable contribution to the binding free
energy and is not fully compensated by the formation of one
salt bridge between the guanidinium and each of the dianions.
Therefore, the binding is more likely to occur by the
formation of the ion pairs between guanidinium and only
one of the phosphates, as opposed to the formation of a
chelate complex as shown in Figure 4. The geometry of the
trans,trans isomer may be more favorable for the formation of
such single ion pairs.
Compounds 1c and 1d failed to undergo any isomer-
izations, either upon direct irradiation or in the presence of
various sensitizers.[19] We assume that a highly conjugated
chalcone structure may prevent the localization of the
absorbed light energy to the double bond which is necessary
for the formation of a perpendicular intermediate of the
isomerization process.[20] Direct irradiation of 1b led only to
the formation of the cis,trans isomer. Sensitized irradiation in
the presence of 2,3-pentanedione resulted in the formation of
a detectable amount of the cis,cis isomer. However, no
complexation of any of the isomers with dodecylguanidinium
hydrochloride in CDCl3 was observed by NMR spectroscopy.
The cis,cis isomer of compound 2b showed marginal
selectivity in binding to guanidinium over the other two
isomers (Table 1). However, both the absolute affinity and the
selectivity of hydrogen bonding in chloroform were quite low.
Similar behavior and comparable affinities were observed for
the isomers of compound 3b. We believe that the modest
affinities of these receptors are due to the steric hindrance
created by methyl substituents at the oxygen atoms that may
serve as the hydrogen-bond acceptors. This effect is partic-
ularly pronounced in compound 3b, in which two methyl
groups at each phosphate apparently reduce their potentially
very strong ability to form hydrogen bonds.[21] There are
probably entropic and stereochemical reasons for the low
binding selectivity among the isomers of 2a and 3a. The
formation of chelate complexes (Figure 4a) with multiple
hydrogen bonds requires a more precise receptor preorgani-
zation than that for electrostatic complexation. Such preor-
ganization in the cis,cis isomers of 2a and 3a may be
disfavored because of both the higher entropy loss upon
complexation and the steric hindrance created by the methyl
groups.
The results of screening show that compound 2a in its
dicationic form possesses the necessary properties to be
applied in our method. The approach was further implement-
ed with the isomers of 2a in ethanol as a solvent, where the
optimal ratio of the solubility/binding properties was ach-
ieved.[22]
One of the important issues we intended to address in this
model study was the necessity for physical separation of the
selection and variation sites. Indeed, one might be able to
achieve the desired self-organization of the isomeric mixture
merely by adding the ligand to the equilibrating system of
receptors, thereby shifting the isomerization equilibrium (see
Figure 3). In order to check this possibility in our system, we
performed long-term irradiation of the receptor solution in
the absence and in the presence of an excess of methylgua-
nidinium hydrochloride. Although the addition of the ligand
to the solution did cause a modest shift of the photostationary
distribution of the isomers (Figure 5), a much stronger effect
Figure 5. Photostationary distribution [%] of isomers of 2a (2.6 Â 10 4 m in
ethanol): a) with no additive; b) in the presence of 0.1m methylguanidinium
hydrochloride.
can be predicted (see below) on the basis of the binding-
constant values. It is likely that direct addition of methyl-
guanidinium affects the equilibrium by mechanisms other
828
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0828 $ 17.50+.50/0
Chem. Eur. J. 1998, 4, No. 5

Automated Genetic Algorithms
825±834
than formation of electrostatic complexes. Possible effects
may include changes in the extinction coefficients and
alteration of the excited-state energies of the receptor isomers
due to the complexation.[23] Any of these effects would cause
intrinsic changes in the photostationary equilibrium and result
in a relatively inefficient enrichment of the system with the
effective binder. This fact justifies separation of the selection
and isomerization sites and may have important implications
for the use of this method in combinatorial chemistry (see below).
Since selection of the effective receptor forms by the
immobilized ligand seemed to have advantages over one-
phase selection, we then explored various sorbents as possible
supports for the guanidinium derivatives.[24] After the sorbents
had been modified with different loadings of protected
arginine (see Experimental Section), they were equilibrated
with an irradiated 3 Â 10 4 m solution of the disodium salt of
2a in ethanol, and the residual distribution of isomers in
solution was analyzed by HPLC.[25] As shown in Table 2, the
modified silica gel was found to be optimal among the tested
supports because of its low nonspecific adsorption, low flow
resistance, and high specific surface. This sorbent and agarose,
Figure 6. HPLC trace of the isomeric mixture of 2a before (broken line)
and after (solid line) it passed through the silica column with 5 Â 10 3 m
immobilized arginine (detection at 270 nm); peaks shown correspond to
cis,cis, cis,trans, and trans,trans isomers, in increasing order of retention
time. A ˆ absorbance.
which the variation compartment was represented by a
photochemical flow cell immersed in a Rayonet reactor
(Figure 7). First, the isomer distribution in the mixture,
reached after the solution had been in
Table 2. Adsorption of the isomers of 2a from ethanol solution on arginine-modified sorbents
(5mm Arg on all supports).
the irradiation cell at the given flow rate
for one radiation cycle, was analyzed by
HPLC (Figure 8a). The time (approx.
5 min) was sufficient to achieve only
partial isomerization; this allowed us to
use higher flow rates, shorten the total
experimental time, and minimize the side-
reactions. The circulation experiments
were then performed, and the composi-
tions of the ligand-bound and the solution-
phase receptors were analyzed as descri-
bed in the Experimental Section.
Support
Isomer distribution in solution
% of bound isomer
(cis,cis/cis,trans/trans,trans, relative peak areas) (cis,cis/cis,trans/
trans,trans)
before column
after column
aminopropyl silica
oxirane acrylic beads 16.8/21.6/8.7
16.5/20.9/8.6
0.8/8.0/5.3
0/0/0
95.32/61.7/37.8
100/100/100
Celite
agarose
Merrifield resin
16.3/20.8/8.2
1/6.1/69.5
8.5/21.9/14.0
0/0/5.3
0/0.59/11.9
0/0/0.03
100/100/35
> 95/90.4/82.8
100/100/ > 99
which showed an appreciable nonspecific adsorption, were
later used for the circulation experiments. More hydrophilic
supports, such as oxirane acrylic beads, Celite, and the
Merrifield resin, absorbed a major amount of all the receptor
forms and were abandoned in further studies.
Using 1 Â 10 2 m arginine immobilized on the silica support
resulted in the distribution of isomers on the column shown in
Figure 8b, which indicated a significant enrichment of the
The concentration of the ligand to be attached to the solid
support for the preferential selection of the receptor isomers
was chosen according to the solution binding constants. Thus,
with 5 Â 10 3 m immobilized arginine the trans,trans isomer
would remain mostly in solution, the cis,trans isomer would be
approximately half-bound to the ligand, and the majority of
the cis,cis isomer would be bound to the solid support.
However, since the solution association constants might not
necessarily be the same for binding to supported arginine, we
tested the distribution of the isomeric mixture between the
solution and the immobilized ligand. Figure 6 shows the
distribution of isomers in the mixture before and after passing
the column with 5 Â 10 3 m ligand immobilized on the silica
support. As predicted, equilibration with the support indeed
led to preferential binding of the effective isomer. In order to
maximize the yield of cis,cis-2a in our system, a somewhat
higher ligand concentration was used for the circulation
experiment.
Figure 7. The experimental setup.
system compared with the photostationary distribution in
solution (cis,cis/cis,trans ratios 6.5:1 and 1.7:1, respectively).
Application of the agarose-supported ligand led to a lower
enrichment (Figure 8c) (cis,cis/cis,trans ratio 2.0:1) due to a
strong nonspecific adsorption (see Table 2).
For performing the selection ± variation cycles, we used an
experimental setup similar to the one shown in Figure 3 in
Chem. Eur. J. 1998, 4, No. 5
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0829 $ 17.50+.25/0
829

A. V. Eliseev and M. I. Nelen
FULL PAPER
ucts.[26] In contrast, the mixture
removed from the silica sup-
port after the evolution experi-
ment (54% of the total original
amount) contained no compo-
nents other than the isomers of
2a (Figure 9b), all of the by-
products having remained in
the circulating solution (Figur-
e 9c).
In order to check whether
the result of evolution was
actually caused by differential
binding to the immobilized
arginine, a control experiment
Figure 8. Distribution [%] of the isomers of 2a in the circulation experiments: a) in 2.6 Â 10 4 m ethanol
solution, after the first irradiation cycle; b) on the arginine-modified silica column, after 30 irradiation ±
selection cycles; c) on the arginine-modified agarose column, after 100 irradiation ± selection cycles.
was performed in which the amino groups on the silica
support were acetylated instead of being modified with
arginine. After 30 cycles with this blank sorbent, only 12%
of the isomers had accumulated on the column, their ratio
(cis,cis/cis,trans/trans,trans ˆ 55:31:14) being close to the pho-
tostationary distribution in solution. After 30 cycles with the
same sorbent but without irradiation, the column distribution
was cis,cis/cis,trans/trans,trans ˆ 5:20:75 (12% total), probably
because of a minor thermal isomerization.
As a positive side-effect of the variation ± selection cycles,
the column-accumulated subset of receptors was free from
any detectable by-products of photochemical isomerization.
Thus, after an 8 h irradiation of 2a in solution, leading to a
photostationary distribution of the components (see Figur-
e 5a), a total of 66% of all receptor forms remained in
solution. HPLC showed the formation of two additional peaks
(Figure 9a), probably attributable to the condensation prod-
The results obtained prove that application of the selec-
tion ± isomerization process to the pool of potential receptors
allows amplification of the total amount of the best receptor
in the reaction setup, and that such an amplification is driven
by the differential affinity of the pool components for the
immobilized target. Notably, the use of the appropriate
immobilization technique makes it possible to maintain the
binding affinity and selectivity observed in solution.[27]
Mathematical modeling of the approach: As shown above, the
approach described here is designed to shift the equilibrium of
the variation process (see Figure 3) to the formation of the
most effective binders by means of their complexation with
the ligand. However, in the implementation of the method,
every variation cycle results in only a limited degree of
transformation of the mixture components and does not bring
the system to equilibrium. This raises important questions,
such as whether the kinetically controlled isomerization
would lead to the optimal evolution of the equilibrating pool,
and whether and how the degree of variation affects the yield
of the effective receptors. In order to address these issues, we
have performed mathematical modeling of the real-time
circulation process and compared the results with both the
experimental data and the calculated equilibrium distribution
of the receptor among the isomers in the free and bound
forms.
Real-time simulation: The first step in the modeling process
involved calculation of the degree of isomerization for the
time the mixture spent in the variation chamber during the
first cycle. The assumption was made that the rate constant for
the transformation of the cis form to the trans form (kc!t) and
the reverse constant (kt!c) are independent of the state of the
receptor (i.e., k(trans,trans !cis,trans) ˆ k(cis,trans !cis,cis)
and k(cis,cis !cis,trans) ˆ k(cis,trans !trans,trans). The val-
Figure 9. HPLC traces of the isomeric mixtures of 2a: a) in solution, after
8 h of irradiation; b) on the arginine-modified silica column, after 30
isomerization ± selection cycles; c) in solution, after 30 cycles. A ˆ absorb-
ance (278 nm).
830
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0830 $ 17.50+.50/0
Chem. Eur. J. 1998, 4, No. 5

Automated Genetic Algorithms
825±834
ues of the constants were determined from the initial rate of
1
isomerization of trans,trans-2a (kt!c ˆ 0.0182 min ) and the
equilibrium constant (Kps) of photostationary distribution of
the isomeric forms (kc!t ˆ kt!c/Kps ˆ kt!c[trans]ps/[cis]ps ˆ
1
0.00874 min , where [trans]ps and [cis]ps are the photosta-
tionary concentrations of the respective forms). By solving the
differential equation for the reversible cis ± trans transforma-
tion, one can find the concentrations of the trans and cis forms
at a particular time point (t) [Eqs. (1) and (2); c0 and t0 are the
initial concentrations of the cis and trans forms, respectively].
1
[cis] ˆ
[kt!c(t0 ‡ c0) ‡ (kc!tc0 kt!ct0)exp( (kt!c ‡ kc!t)t)]
(1)
kt!c ‡ kc!t
1
[trans] ˆ
[kc!t(t0 ‡ c0) ‡ (kt!ct0 kc!tc0)exp( (kt!c ‡ kc!t)t)] (2)
kt!c ‡ kc!t
Figure 10. Calculated distribution of the isomers of 2a in the circulation
The concentrations of the cis and trans forms at the end of
one variation cycle obtained in this way were then used to
express the concentrations of particular isomers by numerical
solution of Equations (3) ± (5).
!
*
setup as a function of cycle number: trans,trans ) in solution and ( ) on
&
&
*
column; cis,trans ( ) in solution and ( ) on column; cis,cis ( ) in solution
!
and ( ) on column.
simulated similarly: the concentrations in solution of the
isomers from the preceding selection were used as the initial
concentrations in the next calculation set of isomerization
kinetics.
The results of modeling 30 variation ± selection cycles are
shown in Figure 10. The final distribution of the isomers on
the column is in very good agreement with the experimental
results. Notably, the system reaches macroequilibrium be-
tween the solution and the solid state through a steady
increase in the cis,cis isomer content and a decrease in the
trans,trans isomer content, whereas the concentration of the
cis,trans isomer passes its maximum, as is typical for the
kinetics of regular consecutive reactions.
2[cis,cis] ‡ [cis,trans] ˆ [cis]
(3)
(4)
(5)
2[trans,trans] ‡ [cis,trans] ˆ [trans]
2 ‰trans; transŠ ‰cis; transŠ
ˆ
‰cis; transŠ
2 ‰cis; cisŠ
These calculations yield the concentrations of isomers in
the solution that leaves the variation chamber and is trans-
ferred to the selection column, which were then used to
calculate the distribution of isomers between the bound and
free forms in the selection column. This was achieved by
finding a numerical solution of three complexation equilibria
[Eqs. (6) ± (8)] and four mass-balance equations, for each of
Equilibrium simulations: The main purpose of these simu-
lations was to compare the results of the circulation experi-
ment with a hypothetical situation, depicted in Figure 2, in
which the ligand introduced into the solution would just shift
the isomerization equilibrium by binding the effective recep-
tor isomers. This system was modeled by taking into account
the equilibrium distribution of the isomers in the photosta-
tionary state [Eqs. (5), (13)] and association of each of them
‰trans; transŠ
col
Ktt ˆ
Kct ˆ
(6)
(7)
(8)
‰LŠ‰trans; transŠ
sol
‰cis; transŠ
col
‰LŠ‰cis; transŠ
sol
‰cis; cisŠ
col
Kcc
ˆ
‰LŠ‰cis; cisŠ
sol
the isomers [Eqs. (9) ± (11)] and for the immobilized ligand
[Eq. (12)] ([L] and [L]t are the receptor-unbound and total
concentrations of immobilized arginine, respectively; sub-
2 ‰cis; cisŠ ‡ ‰cis; transŠ
sol
sol
ˆ Kps
sol
(13)
2 ‰trans; transŠ ‡ ‰cis; transŠ
sol
[trans,trans] ˆ [trans,trans]col ‡ [trans,trans]sol
[cis,trans] ˆ [cis,trans]col ‡ [cis,trans]sol
[cis,cis] ˆ [cis,cis]col ‡ [cis,cis]sol
(9)
(10)
(11)
(12)
with the ligand [Eqs. (6) ± (8)]. Solution of this system to-
gether with the corresponding mass-balance equations for
our experimental conditions yielded the column distribution
shown in Figure 11a, in which the component concentrations
exactly matched the level-off values for the components in
Figure 10. Similar calculations were also performed taking
into account the (80%) nonspecific adsorption of the agar-
ose-supported ligand (Figure 11b). The distribution obtained
also corresponds to the one observed experimentally (cf.
Figure 8c).
[L]t ˆ [L] ‡ [trans,trans]col ‡ [cis,trans]col ‡ [cis,cis]col
scripts col and sol refer to the concentrations of column-
bound and free receptor forms, respectively). The resulting
array of concentrations is the result of the first cycle, as
shown in the first set of points (cycle 1) in Figure 10. The
second and following cycles of isomerization ± selection were
These results indicate that the enrichment of the pool
obtained after a sufficient number of cycles in our exper-
imental setup is essentially driven by the shift of the variation
Chem. Eur. J. 1998, 4, No. 5
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0831 $ 17.50+.25/0
831

A. V. Eliseev and M. I. Nelen
FULL PAPER
Table 3. Simulation of the evolutionary enrichment of multicomponent
mixtures by the selection ± variation method.
No. of
Kstrong/Kweak
Column yield of the Total yield of strong
components
strong binder [%]
binder [%]
5
5
5
10
10
10
15
15
15
5
10
50
10
50
100
15
50
55.6
71.4
92.6
52.6
84.7
91.7
51.7
78.1
87.7
34.9
51.6
84.0
33.7
71.5
83.3
33.3
62.3
76.7
Figure 11. Calculated distribution [%] of isomers of 2a: a) on the arginine-
modified silica column; b) on the arginine column (cf. Figures 8b, 8c,
respectively).
100
equilibrium by the immobilized ligand. One of the important
conclusions resulting from these simulations is that the final
result of circulation is independent of the degree of isomer-
ization in each cycle. Thus, just a minor mutation performed in
the isomerization chamber would eventually lead to the same
distribution of the components that would be achieved by
bringing the system to the isomerization equilibrium every
time.[28] This conclusion is essential for the application of the
method to more complex pools and slow variation reactions.
The final total ratio of the isomers in the experimental setup
corresponds to their thermodynamic distribution in the
situation when the ligand is initially dissolved in the irradiated
receptor solution, but there are no side-reactions involved.
Therefore, the final evolution result, even in the system with
kinetically controlled variation, is determined by the thermo-
dynamics of the interconversion equilibrium and the ligand ±
receptor interactions. Since this observation sets an upper
limit on the enrichment that can be achieved by our method, it
was interesting to use the simulations described above to
evaluate the applicability of the approach to larger pools.
We considered three sample pools, containing 5, 10, and 15
components respectively, and estimated how the affinity
difference between the components would influence the
effectiveness of selection by the ligand. For simplicity, all of
the pool components were assumed to have equal binding
constants to the ligand (Kweak), except for one that was
assigned a higher constant (Kstrong). The ligand concentration
was chosen to equal 1/Kweak, in order to leave half of each
weak binder in solution and, at the same time, to bind the
majority of the effective component. The equilibrium calcu-
lations, similar to those described above for the three-
component mixture, were then performed for different Kstrong
values. The results of these simulations were expressed in
terms of yield of the strong binder on the column, defined as
the ratio (amount of strong binder on column)/(total amount
of receptor on column), and the total yield of the strong
binder, defined as (amount of strong binder on column)/(total
amount receptor in the experimental setup) (Table 3). As
expected, the efficiency of the method is highly dependent on
the selectivity of ligand recognition by different receptor
forms. The increasing complexity of the calculations did not
allow us to evaluate pools larger than 15 components. The
general trend shows, however, that the mixture of n compo-
nents in which Kstrong/Kweak ˆ n would be evolved by repeating
the selection ± equilibration to afford approximately 50% of
the effective component on the column, its total yield being
approximately 30%. Such a situation seems to be realistic for
combinatorial libraries where the hits may possess a target
affinity several orders of magnitude higher than the majority
of the components.[29]
Conclusions and Outlook
In this paper we have assessed the mechanism, applicability,
and possible extensions of the method of targeted evolution of
equilibrating pools driven by the differential affinity of the
pool components for a chosen target compound. In a way, this
method represents a thermodynamic analogue of template
synthesis[30] in which assembly of the reactants on the
template leads to a faster coupling of the preassembled units
and thereby eliminates the formation of multiple side-
products that could have been formed in the absence of a
template. It is important, however, to emphasize the differ-
ences between evolutionary selection ± variation and the
traditional template-directed synthesis. Thermodynamic con-
trol of the selection process ensures that the compounds
accumulated on the support are indeed the best binders for
the given target. Also, the template (or target) in our method
does not interfere with the variation reaction, however severe
the conditions that are used.
The latter point is particularly important for the applic-
ability of the technique in combinatorial chemistry, where the
targets, such as proteins or nucleic acids, may be particularly
fragile. The method may be especially promising for combi-
natorial chemistry in that it allows one to combine the
generation of diversity and screening in one system that, in
addition, can be easily automated. Generation of larger
equilibrating pools, which is a focus of our current work, can
be based on many known types of reversible reactions, such as
imine exchange, transesterification, thiol ± disulfide exchange,
and aldol condensations, to name just a few.
Experimental Section
General: Reagents for synthesis and compound 1a (the diacid form) were
obtained from commercial sources and used without further purification.
NMR spectra were obtained and processed on Varian Unity 300 and
400 MHz spectrometers; HPLC analysis was performed on the Beckman
Gold system equipped with a Model 168 photodiode array detector.
832
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0832 $ 17.50+.50/0
Chem. Eur. J. 1998, 4, No. 5

Automated Genetic Algorithms
825±834
Mathcad Pro software (version 6.0) running on a Power Macintosh 6100
was used for the numerical simulations.
from the changes in chemical shifts (Dd) of the host signals by nonlinear
curve fitting to Equation (14) for the 1:1 complexation mode.
Synthesis of phenylenediacrylic acid pyrazolylamide (1b): Phenylenedi-
acrylic acid (1a) (2 g, 9.17 mmol) was transformed into its dichloroanhy-
dride by refluxing overnight in SO2Cl2/CH2Cl2 (40 mL, 2m) with a catalytic
amount of DMF, followed by evaporation of the solvent. The dichlor-
oanhydride (1.17 g, 4.6 mmol) was dispersed in anhydrous Et2O (30 mL)
containing pyrazole (0.750 g, 11 mmol), Et3N (1.53 mL, 11 mmol), and 4-
dimethylaminopyridine (DMAP)(20 mg), and heated to reflux for 18 h.
The off-white precipitate was then filtered, washed twice with Et2O,
dissolved in CHCl3 (150 mL), extracted three times with water, and dried
over MgSO4. Crystallization from CHCl3/hexanes yielded 1b (833 mg,
Dd ˆ Dd K[G]/(1 ‡ K[G])
(14)
1
With a few exceptions, the variations of the binding constants determined
from the shifts of different protons did not exceed the experimental error
(approx. 10 ± 15%).
Ligand immobilization: N-(a-tert-Butoxycarbonyl)arginine was attached to
the solid support by carbodiimide coupling with chromatographic sorbents
modified with amino groups (aminopropyl silica was prepared by a
procedure described in ref. [32]; other supports were commercially
available). In a typical procedure (for silica), the amino sorbent (4 g,
6.7 mL dry volume) was dispersed in dry DMF (15 mL); the solutions of the
arginine derivative (10.5 mg, 0.0335 mmol) in DMF (2 mL) and 1-ethyl-3-
(3-dimethylaminopropyl)carbodiimide (EDC) (89 mg, 0.3 mmol) in DMF
(2 mL) were added. The resulting slurry was shaken for 18 h on a device
with a wrist-shaking action. Then Ac2O (2 mL, 21.2 mmol) and Et3N (3 mL,
22 mmol) were added and the mixture was shaken for another 24 h, then
filtered, and washed successively with DMF, water, and ethanol.
1
57%). H NMR (300 MHz, CDCl3): d ˆ 8.41,8.40 (d, 2H), 8.07, 8.01, 7.99,
7.94 (dd, 4H), 7.81 (s, 2H), 7.76 (s, 4H), 6.52 (m, 2H); 13C NMR (75 MHz,
CDCl3): d ˆ 163.9, 146.9, 144.6, 137.3, 129.9, 129.3, 117.7, 110.6.
Synthesis of 3,3'-(1,4-phenylene)-bis[1-(2-thiazolyl)-2-propen-1-one] (1c):
To a solution of terephthalaldehyde (0.134 g, 1.0 mmol) and 2-acetyl-
thiazole (0.254 g, 2.0 mmol) in MeOH (5 mL) was added saturated NaOH
in MeOH (0.1 mL). The mixture was stirred for 3 h at room temperature,
then the precipitate formed was filtered off, washed with MeOH, dried, and
crystallized from EtOAc to afford 1c (110 mg, 31%). 1H NMR (300 MHz,
CDCl3): d ˆ 8.10,8.09 (d, 2H), 8.02 (s, 4H), 7.78 (s, 4H), 7.74,7.73 (d, 2H).
Calcd for C18H12O2N2S2: C, 61.3, H, 3.43, N, 7.95; found: C, 60.8, H, 3.55, N,
7.79.
Circulation experiments: The circulation experiments were performed at a
typical flow rate of 0.5 ± 1 mLmin for 8 ± 12 h, resulting in 30 ± 100 cycles.
1
The transparent flow cell made from borosilicate glass was irradiated with
broad-band UV± visible light in a Rayonet reactor; the same cell was used
to determine the photostationary distribution in solution. The experiments
were terminated when the HPLC analysis of the circulating solutions
showed no further changes in the concentrations of the components in
solution. When cycling had been terminated, the sorbent removed from the
column was treated with several portions of 1m NaCl in water and aqueous
ethanol. Combined extracts were concentrated in vacuo and analyzed by
HPLC as described in ref. [9].
Synthesis of 3,3'-(1,4-phenylene)-bis(1-(2-pyridyl)-2-propen-1-one) (1d):
The same procedure as that for 1c was used with 2-acetylpyridine instead of
2-acetylthiazole. Yield 46%. 1H NMR (300 MHz, CDCl3): d ˆ 8.78 ± 8.77,
8.77 ± 8.76 (dq, 2H), 8.39, 8.34 (d, 2H), 8.22, 8.20 (dt, 2H), 7.97, 7.92 (d,
4H), 7.92, 7.90, 7.87 (td, 2H), 7.79 (s, 4H), 7.54 ± 7.49 (qd, 2H); 13C NMR
(75 MHz, CDCl3): d ˆ 189.8, 154.7, 149.4, 144.1, 137.8, 137.6, 129.8, 127.5,
123.5, 122.5.
The dicarboxylic acid 2a and its dimethyl ester 2b were prepared and
characterized as described in ref. [18].
4-(4-Formylphenoxy)benzaldehyde: This intermediate, used in the syn-
thesis of 3b, was prepared by the reaction of 4-hydroxybenzaldehyde with
4-fluorobenzaldehyde as described previously.[31] Yield 89%. 1H NMR
(300 MHz, CDCl3): d ˆ 9.98 (s, 2H), 7.95, 7.20 (dd, J ˆ 9.0 Hz, 8H).
Acknowledgments: This work was supported by a start-up grant to A.V.E.
from SUNY/Buffalo. We also acknowledge the Donors of the Petroleum
Research Fund administered by the American Chemical Society for partial
support of this research.
Tetramethyldiphosphonate 3b:
A solution of tetramethylmethylene-
diphosphonate (1 g, 4.3 mmol) in anhydrous THF (2 mL) was added
dropwise to a stirred suspension of NaH (0.099 g, 4.1 mmol) in THF (2 mL)
at 08C in an argon atmosphere. The mixture was adjusted to room
temperature and kept for 40 min until the solution became transparent and
no further gas formation was observed. A solution of 4-(4-formylphen-
oxy)benzaldehyde (0.406 g, 1.8 mmol) in THF (2 mL) was added dropwise
within 30 min at 10 ± 158C, and the resulting mixture was stirred for 2 h at
ambient temperature. Then the solvent was removed in vacuo, and the
residue was redissolved in Et2O, extracted twice with water and then with a
saturated solution of NaCl, and dried over MgSO4. The crude product was
purified by chromatography on silica gel in 15% (v/v) MeOH in EtOAc,
yielding the analytically pure product (0.380 g, 48%). 1H NMR (300 MHz,
CDCl3): d ˆ 7.57, 7.52, 7.50, 7.44 (q, 2H), 7.52, 7.50, 7.05, 7.02 (dd, 8H), 6.20,
6.14, 6.09 (t, 2H), 3.80, 3.77 (d, 12H); 31P NMR (160 MHz, CDCl3): d ˆ
22.8.
Received: July 10, 1997 [F762]
[1] a) Genetic Algorithms and Simulated Annealing (Ed.:L. Davis),
Morgan Kaufmann, London, Los Altos, 1987; b) J. H. Holland, Sci.
Am. 1992, 66 ± 72; c) S. Forrest, Science 1993, 261, 872 ± 878.
[2] For reviews on application of GAs to chemical problems, see: a) G.
Quinkert, H. Bang, D. Reichert, Helv. Chim. Acta 1996, 79, 1260 ±
1278; b) P. Willett, Trends Biotechnol. 1995, 13, 516 ± 521; c) D. E.
Clark, D. R. Westhead, J. Computer-Aided Mol. Design 1996, 10, 337 ±
358.
[3] a) L. Weber, S. Wallbaum, C. Broger, K. Gubernator, Angew. Chem.
1995, 107, 2452 ± 2454; Angew. Chem. Int. Ed. Engl. 1995, 34, 2280 ±
2282; b) J. Singh, M. A. Ator, E. P. Jaeger, M. P. Allen, D. A. Whipple,
J. E. Soloweij, S. Chowdhari, A. M. Treasurywala, J. Am. Chem. Soc.
1996, 118, 1669 ± 1676; c) Y. Yokobayashi, K. Ikebukuro, S. McNiven,
I. Karube, J. Chem. Soc. Perkin Trans. 1, 1996, 2435 ± 2437.
[4] a) M. Famulok, J. W. Szostak, Angew. Chem. 1992, 104, 1001 ± 1011;
Angew. Chem. Int. Ed. Engl. 1992, 31, 979 ± 988; b) J. R. Lorsch, J. W.
Szostak, Acc. Chem. Res. 1996, 29, 103 ± 110; c) J. Ciesiolka, M.
Illangasekare, I. Majerfeld, T. Nickles, M. Welch, M. Yarus, S. Zinnen,
Methods Enzymol. 1996, 267, 315 ± 335, and other reviews in the same
volume.
Diphosphonate 3a: 3b (0.056 g, 0.128 mmol) was dissolved in dry CH2Cl2
(7 mL), and TMS bromide (0.1 mL, 0.76 mmol) was added at 08C. After
this mixture had been stirred for 10 min, the temperature was adjusted to
ambient, and stirring was continued for 5 h. Then the solvent was removed;
the dry residue was redissolved in MeOH (7 mL), kept overnight, and then
heated to reflux for another 2.5 h. Removal of the solvent in vacuo resulted
in analytically pure 3a (48 mg, 98%). 1H NMR (300 MHz, [D6]DMSO):
d ˆ 7.66, 7.64 (d, 4H), 7.25, 7.19, 7.18, 7.12 (q, 2H), 7.06, 7.04 (d, 4H), 6.47,
6.42, 6.36 (t, 2H); 31P NMR (160 MHz): d ˆ 13.8.
NMR titrations: Solutions of the trans,trans forms of the receptors (1a and
2a as their disodium salts, and 3a as the di(tetrabutylammonium) salt) in
appropriate deuterated solvents were irradiated in NMR tubes with a
mercury lamp in a Rayonet photochemical reactor for 20 ± 30 min to
generate mixtures of all the isomers at a total concentration of 0.5 ± 4mm;
these were titrated with the chloride of the corresponding guest cation G,
with an increasing concentration of the latter from [G] ˆ 0 to 130mm. The
individual binding constants K of all the isomers were then determined
[5] P. Kast, D. Hilvert, Pure Appl. Chem. 1996, 68, 2017 ± 2024, and
references therein.
[6] For reviews, see: a) B. G. Bag, G. von Kiedrowski, Pure Appl. Chem.
1996, 68, 2145 ± 2152; b) L. E. Orgel, Nature (London) 1992, 358,
203 ± 209; c) L. E. Orgel, Acc. Chem. Res. 1995, 28, 109 ± 118;
d) E. A. Wintner, M. M. Conn, J. Rebek, Jr., ibid. 1994, 27, 198;
e) E. A. Wintner, J. Rebek, Jr., Acta Chem. Scand. 1996, 50, 469 ± 485.
[7] a) P. A. Bachmann, P. Walde, P. L. Luisi, J. Lang, J. Am. Chem. Soc.
1991, 113, 8204 ± 8209; b) P. A. Bachmann, P. L. Luisi, J. Lang, Nature
Chem. Eur. J. 1998, 4, No. 5
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0833 $ 17.50+.25/0
833

A. V. Eliseev and M. I. Nelen
FULL PAPER
(London) 1992, 357, 57 ± 59; c) R. Maoz, S. Matlis, E. DiMasti, B. M.
Ocko, J. Sagiv, ibid. 1996, 384, 150 ± 153.
[18] M. I. Nelen, A. V. Eliseev, J. Chem. Soc. Perkin Trans. 2 1997, 1359 ±
1364.
[8] a) K. Soal, T. Shibata, H. Morioka, K. Choji, Nature (London) 1995,
378, 767 ± 768; b) T. Shibata, H. Morioka, T. Hayase, K. Choji, K.
Soai, J. Am. Chem. Soc. 1996, 118, 471 ± 472; c) C. Bolm, F.
Bienewald, A. Seger, Angew. Chem. 1996, 107, 1767 ± 1769; Angew.
Chem. Int. Ed. Engl. 1996, 35, 1657 ± 1659.
[19] The sensitizers were chosen to cover the triplet energy (ET) range of
45 ± 69 kcalmol 1, from the list given in: S. L. Murov, I. Carmichael,
G. L. Hug, Handbook of Photochemistry, 2nd ed., Marcel Dekker,
New York, 1993.
[20] For a description of the mechanisms of photochemical cis ± trans
isomerization, see: a) G. S. Hammond, J. Saltiel, A. A. Lamola, N. J.
Turro, J. S. Bradshaw, D. O. Cowan, R. C. Counsell, V. Vogt, C.
Dalton, J. Am. Chem. Soc. 1964, 86, 3197 ± 3217; b) N. J. Turro,
Photochem. Photobiol. 1969, 9, 555 ± 563. For the isomerization of
chalcones, see: c) M. Reinkhardt, V. G. Mitina, N. S. Pivenko, V. F.
Lavrushin, Zh. Obshch. Knim. 1980, 50, 2770 ± 2776.
[9] A. V. Eliseev, M. I. Nelen, J. Am. Chem. Soc. 1997, 119, 1147 ± 1148.
[10] For representative reviews on combinatorial chemistry and the
principles of library design, see: a) M. A. Gallop, R. W. Barrett,
W. J. Dower, S. P. A. Fodor, E. M. Gordon, J. Med. Chem. 1994, 37,
1233 ± 1251; b) E. M. Gordon, R. W. Barrett, W. J. Dower, S. P. A.
Fodor, M. A. Gallop, ibid. 1994, 37, 1385 ± 1401; c) N. K. Terrett, M.
Gardner, D. W. Gordon, R. J. Kobylecki, J. Steele, Tetrahedron 1995,
51, 8135 ± 8173; d) J. Eichler, R. A. Houghten, Mol. Med. Today 1995,
1, 174 ± 180; e) G. Lowe, Chem. Soc. Rev. 1995, 24, 309; f) L. A.
Thompson, J. A. Ellman, Chem. Rev. 1996, 96, 555 ± 600; g) X.
Williard, I. Pop, L. Bourel, D. Horvath, R. Baudelle, P. Melnyk, B.
Deprez, A. Tartar, Eur. J. Med. Chem., 1996, 31, 87 ± 98; h) F.
Balkenhohl, C. von dem Bussche-Hünnefeld, A. Lansky, C. Zechel,
Angew. Chem. 1996, 108, 2436 ± 2487; Angew. Chem. Int. Ed. Engl.
1996, 35, 2288 ± 2337.
[11] In an example of such a target-promoted equilibrium shift described
earlier (W. C. Still, P. Hauck, D. Kempf, Tetrahedron Lett. 1987, 28,
2817 ± 2820), the natural ionophore antibiotic lasalocid A was formed
by epimerization of its stereoisomer in the presence of a barium salt.
[12] a) D. L. Venton, A. J. Hopfinger, G. Le Breton, US 5366862, 1994;
b) P. G. Swann, R. A. Casanova, A. Desai, M. M. Frauenhoff, M.
Urbancic, U. Slomczynska, A. J. Hopfinger, G. C. Le Breton, D. L.
Venton, Biopolymers 1996, 40, 617 ± 625.
[21] M. H. Abraham, Chem. Soc. Rev. 1993, 22, 73 ± 83.
[22] Compound 1a,which showed binding properties similar to those of 2a
in methanol, was poorly soluble in ethanol and less-polar media where
the required binding selectivity would be expected.
[23] D. C. Neckers, Mechanistic Organic Photochemistry, Reinhold, New
York, 1967, p. 199.
[24] The limited solubility of arginine derivatives in ethanol and other
organic solvents prevented us from using them directly for the solution
equilibrium studies. However, a derivative of the authentic biological
compound was used for immobilization on the solid support.
[25] The irradiated solution of 2a was slowly pumped through the sorbent-
filled column, and the content of isomers was determined in the
emerging fractions (a total of 2 ± 3 column volumes).
[26] a) Y. Nakamura, J. Chem. Soc. Chem. Comm. 1982, 477; b) M.
DꢀAuria, A. Vantaggi, Tetrahedron 1992, 48, 2523.
[27] The technique of multiple isomerization ± isolation steps has been
used previously, in a system of porphyrin isomers, to increase the yield
of the desired form: a) J. Lindsey, J. Org. Chem. 1980, 42, 5215;
b) C. M. Elliott, Anal. Chem. 1980, 52, 666 ± 668. We thank Prof.
Lindsey for making us aware of this work. Our approach, however,
aims at the generation of molecular entities capable of strong and
specific binding to target compounds in solution, for which the
similarity of binding between the immobilized and the free ligand is
essential.
[28] Apparently, the smaller the degree of isomerization, the more cycles
are necessary for achieving the macroequilibrium. The total time for
macroequilibration is, however, comparable with the time necessary
to reach a similar solution equilibrium and can be estimated from
straightforward kinetic studies in solution.
[29] a) S. M. Freier, D. A. M. Konings, J. R. Wyatt, D. J. Ecker, J. Med.
Chem. 1995, 38, 344 ± 352; b) L. Wilson-Lingardo, P. W. Davis, D. J.
Ecker, N. Hebert, O. Acevedo, K. Sprankle, T. Brennan, L. Schwarsz,
S. M. Freier, J. R. Wyatt, J. Med. Chem. 1996, 39, 2720 ± 2726.
[30] R. Hoss, F. Vögtle, Angew. Chem. 1994, 106, 389 ± 398; Angew. Chem.
Int. Ed. Engl. 1994, 33, 375 ± 84.
[13] I. Huc, J.-M. Lehn, Proc. Natl. Acad. Sci. USA 1997, 94, 2106 ± 2110.
[14] a) S. J. Rowan, P. A. Brady, J. K. M. Sanders, Angew. Chem. Int. Ed.
Engl. 1996, 35, 2143 ± 2145; b) S. J. Rowan, D. G. Hamilton, P. A.
Brady, J. K. M. Sanders, J. Am. Chem. Soc. 1997, 119, 2578 ± 2579;
c) S. J. Rowan, J. K. M. Sanders, Chem. Commun. 1997, 1407 ± 1408.
Â
[15] J. Turkova, Affinity Chromatography , Elsevier, Amsterdam, 1978.
[16] For recent studies on synthetic receptors involved in recognition,
transport, and sensing of guanidinium and amidinium derivatives, see:
a) T. W. Bell, J. Liu, Angew. Chem. 1990, 102, 931 ± 933; Angew.
Chem. Int. Ed. Engl. 1990, 29, 923 ± 925; b) T. W. Bell, V. J. Santora, J.
Am. Chem. Soc. 1992, 114, 8300 ± 8302; c) T. W. Bell, Z. Hou, S. C.
Zimmerman, P. A. Thiessen, Angew. Chem. 1995, 107, 2321 ± 2324;
Angew. Chem. Int. Ed. Engl. 1995, 34, 2163 ± 2165; d) A. Casnati, P.
Minari, A. Pochini, R. Ungaro, W. F. Nijenhuis, F. De Jong, D. N.
Reinhoudt, Israel J. Chem. 1992, 32, 79 ± 87; e) F. J. B. Kremer, G.
Chiosis, J. F. J. Engbersen, D. N. Reinhoudt, J. Chem. Soc. Perkin
Trans. 2 1994, 677 ± 681; f) M. Takeshita, S. Shinkai, Chem. Lett. 1994,
1349 ± 1352.
[17] In a few cases, the receptors isolated after synthesis contained up to
10% of the corresponding cis,trans isomers.
[31] O. Dann, J. Ruff, H.-P. Wolf, H. Griessmeier, Lieb. Ann. Chem. 1984, 409.
[32] R. Wu, L. Grossman, Methods Enzymol. 1987, 154, 299.
834
ꢁ WILEY-VCH Verlag GmbH, D-69451 Weinheim, 1998
0947-6539/98/0405-0834 $ 17.50+.50/0
Chem. Eur. J. 1998, 4, No. 5