Training a Constitutional Dynamic Network for Effector Recognition: Storage, Recall, and Erasing of Information

Article abstract of DOI:10.1021/jacs.6b05785

Constitutional dynamic libraries (CDLs) of hydrazones, acylhydrazones, and imines undergo reorganization and adaptation in response to chemical effectors (herein metal cations) via component exchange and selection. Such CDLs can be subjected to training by exposition to given effectors and keep memory of the information stored by interaction with a specific metal ion. The long-term storage of the acquired information into the set of constituents of the system allows for fast recognition on subsequent contacts with the same effector(s). Dynamic networks of constituents were designed to adapt orthogonally to different metal cations by up- and down-regulation of specific constituents in the final distribution. The memory may be erased by component exchange between the constituents so as to regenerate the initial (statistical) distribution. The libraries described represent constitutional dynamic systems capable of acting as information storage molecular devices, in which the presence of components linked by reversible covalent bonds in slow exchange and bearing adequate coordination sites allows for the adaptation to different metal ions by constitutional variation. The system thus performs information storage, recall, and erase processes.

Full text of DOI:10.1021/jacs.6b05785

Article
pubs.acs.org/JACS
Training a Constitutional Dynamic Network for Effector Recognition:
Storage, Recall, and Erasing of Information
Jan Holub, Ghislaine Vantomme,† and Jean-Marie Lehn*
Laboratoire de Chimie Supramolec
́ ́ ́ ́
ulaire, Institut de Science et d’Ingenierie Supramoleculaires (ISIS), Universite de Strasbourg, 8
allee Gaspard Monge, 67000 Strasbourg, France
́
S
* Supporting Information
ABSTRACT: Constitutional dynamic libraries (CDLs) of
hydrazones, acylhydrazones, and imines undergo reorganiza-
tion and adaptation in response to chemical effectors (herein
metal cations) via component exchange and selection. Such
CDLs can be subjected to training by exposition to given
effectors and keep memory of the information stored by
interaction with a specific metal ion. The long-term storage of
the acquired information into the set of constituents of the
system allows for fast recognition on subsequent contacts with
the same effector(s). Dynamic networks of constituents were
designed to adapt orthogonally to different metal cations by
up- and down-regulation of specific constituents in the final
distribution. The memory may be erased by component
exchange between the constituents so as to regenerate the
initial (statistical) distribution. The libraries described represent constitutional dynamic systems capable of acting as information
storage molecular devices, in which the presence of components linked by reversible covalent bonds in slow exchange and
bearing adequate coordination sites allows for the adaptation to different metal ions by constitutional variation. The system thus
performs information storage, recall, and erase processes.
through (repetitive) application of an effector for the
recognition of that very effector. They comprise (i) imprinting
of information in a constitutional state/distribution of a DCL
by an effector; (ii) long-term retention of the stored
information on removal of the effector; (iii) fast repetitive
recognition of the agent; (iv) retraining of the system onto a
different agent; and (v) final erasing of information with return
to the initial “naive” DCL composition. For the sake of
demonstration, the systems described below implement DCLs
of ligand molecules, responding to the application of metal
cation effectors, based on components related to those used in
previous studies of metallo-selection and photoselection
processes,5 but adjusted here for the present purposes (Figure
1). The results are meant to provide a proof of principle of the
ability of CDC to give access to higher levels of behavior of a
chemical system6 such as training and learning through
adaptation by component selection.3,5
INTRODUCTION
■
The designed self-organization of well-defined supramolecular
architectures from their components rests on the storage of the
required molecular information in the covalent framework of
the components and its processing at the supramolecular level
through specific interactional algorithms.1,2 A further step
toward systems of increasing complexity consists of the
implementation of constitutional dynamic chemistry (CDC)3
and its adaptive features for the design of chemical learning
systems that are not just programmed beforehand for a given
process but can be trained toward the acquisition of a given
function, such as remembering an agent to which it had been
subjected. Training and learning involve progressive adaptation
of a system, in effect a constitutional dynamic library (CDL),
subjected to repetitive application of a physical stimulus or of a
chemical effector, thus building up an adapted state and
retaining it.3d The members of the CDL are linked in a
constitutional dynamic network (CDN)3c,d that interconnects
the constituents of the CDL through agonistic and antagonistic
relationships, so that training and learning imply a progressive
modification/adaptation of the connectome in response to
various inputs/solicitations.
Rationale. The DCL training operations are accessible via
changes between agonist/antagonist pairs of constituents
3
within the CDN. Antagonists, that is, constituents (e.g., A1C
and 3A1D, Figure 1) that share a component (3A), are
competitors. Enhancing the formation of one (3A1C) results in
the reduction of the other (3A1D). On the other hand, agonists
Herein, we demonstrate the implementation of a dynamic
covalent library (DCL)4 for performing a set of molecular
information processing operations involved in the design of a
prototypical chemical learning system, training the system
Received: June 6, 2016
© XXXX American Chemical Society
A
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Figure 1. Structures of the dynamic covalent ligands forming the constitutional dynamic library (CDL) used in the present work. The fragments A,
on one hand, and B, C, and D, on the other, are derived respectively from the aldehyde and the amine (hydrazine, top, or acylhydrazone, middle)
components of the CDL. The complexes of the metal cations M are also shown (bottom).
(e.g., 3A1C and 4A1D, Figure 1) are constituents without
common component. They are in a synergistic relationship;
thus an increase of one (3A1C) enforces also the formation of
the other one (4A1D) by agonist amplification.3c,d
memory/information erasing. Such an effect may be described
as agonist inhibition because, in the presence of both fast
exchanging (imines) and slow exchanging constituents
(hydrazones, acylhydrazones), the stability of one constituent
also prevents re-equilibration of its kinetically less stable
agonist.
The agonist/antagonist relationship in response to inter-
action with an effector, here a metal cation, is the key regulatory
mechanism allowing the DCL to adjust its equilibrium
composition by up- and down-regulation of its constituents.3−5
Exposing the system to a second contact with the same effector
leads to instant recognition in a fast recall process. Such a DCL
can also be amenable to double (multiple) training toward
different effectors (metal cations), and the information stored
can be later erased through re-equilibration of the components.
Depending on the conditions used, orthogonal switching
between different constitutional states through metalloselection
imprints different information represented by specific distribu-
tions of the constituents of the DCL enforced by a given
effector/metal cation. It is important to note that in the present
case the chemical information resides in a distribution of
constituents, representing a constitutional engram,7,8 rather
than in a molecular shape/structure imprint in polymeric
materials.9 Furthermore, removal of the effector leads to an out-
of-equilibrium constitutional state (see also ref 10) that retains
the information about the effector. Memory corresponds to an
out-of-equilibrium state. For the present purposes, the system
has to combine (i) the ability to equilibrate the constituents of
the DCL so as to be able to respond to the effector(s) by
component exchange, (ii) with sufficient stability, that is, slow
component exchange in given conditions, so as to retain the
constituent distribution after removal of the effector(s). At least
one constituent must present these features so as to allow for
kinetic control of the system. Thus, imine-based constituents
exchange too easily their components. On the other hand,
hydrazone- and acylhydrazone-based constituents/ligands are
much more reluctant to undergo component exchange, while
still able to respond to an effector/metal cation. The presence
of an hydrazone or acylhydrazone is thus of utmost importance
for providing the kinetic barrier required for stabilizing the out-
of-equilibrium states generated after removal of the effector/
metal cation and thus preclude re-equilibration, that is,
On the basis of previous work from our group,5 pyridyl-
hydrazones and pyridyl-acylhydrazones present the very
attractive feature of being triple dynamic entities, capable of
long-term and short-term information storage,5a by undergoing
(1) conformational dynamics by shape switching on cation
coordination to the NNN and NNO tridentate binding sites,
respectively;5,11,12 (2) configurational dynamics on E to Z
photoisomerization stabilized by internal hydrogen bonding
with the pyridine moiety;5,13 and (3) constitutional dynamics,
by component exchange via the reversible CN connection.4,5
The DCLs of pyridyl-hydrazones, pyridyl-acylhydrazones, and
pyridyl-imines studied here form networks represented as 2D
square [2 × 2] CDNs in the schemes below. All of the metal
cation-induced changes in constituent distributions were
1
monitored by H NMR spectroscopy and are shown in the
Supporting Information.
RESULTS AND DISCUSSION
■
Training of a Dynamic Covalent Library of Hydra-
zones toward a Single Effector. Imprinting and Storage of
Information. On the basis of previous results in our
laboratory,5 we first investigated the DCL of the four
1
1
2
2
hydrazones A1B, A2B, A1B, and A2B (Figure 1). It was
generated from a mixture of equimolar quantities of the two
1
2
constituents A1B and A2B (0.6 × 10−5 mol, 10 mM each,
through component randomization under metal ion catalysis
(4:1 acetonitrile/water, 170 °C, MW, 20 min in the presence of
Sc(OTf)3 (2 mM)),14 giving the full set of four constituents by
component exchange with
a distribution for
1A1B/1A2B/2A1B/2A2B in agreement within experimental
error with the previously reported one (32/17/21/18%,
respectively, and 12% hydrolysis).5b The A1B was present in
1
the form of its E (60%) and Z (40%) isomers, and the total
B
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
1
1
2
2
Scheme 1. Training of the Dynamic Covalent Library (DCL) of Four Hydrazone Constituents A1B, A2B, A1B, and A2B for
the Recognition of Zinc Cations
a
a
(1) Generation of the complete set of four hydrazones by component exchange under Sc(III) metal ion catalysis and distribution of the
constituents (below); (2) dynamic imprinting by metalloselection via the prime addition of zinc(II) cations and resulting distribution of the
constituents (below) showing adaptation driven by the metal ion effector; (3) removal of both Zn(II) and Sc(III) cations by precipitation with
potassium ferricyanide, with generation of an out-of-equilibrium state storing the information about the Zn(II) effector in the constitution of the
system; (4) new addition of zinc(II) cations inducing a fast recall of the encoded information; and (5) erasing of the stored information. For clarity,
only the major components are represented for each step. (bottom) Representation of the library adaptation as a square Constitutional Dynamic
1
Network (CDN) showing agonist amplification and down-regulation of the antagonists. A1B is obtained as both E and Z (not shown) isomers;
Zn(II) binding converts all Z into E. Error on % determination: ∼ 2%.
amount of both isomers was taken into an account when
analyzing the mixture.
kinetically trapped out-of-equilibrium state, as equilibration is
precluded by the very slow component exchange.
Metalloselection was used to imprint constitutional informa-
tion into the system by changing the relative amounts of the
constituents. Upon treatment with zinc(II) triflate, the DCL
above underwent reorganization to a distribution of 14/29/38/
11% for 1A1B/1A2B/2A1B/2A2B, respectively, with 8% of
hydrolysis, 1A2B being present as the complex Zn(1A2B)2,
and no mixed-ligand complexes were observed (see step 2 in
Scheme 1, Figure 1, and Figure S27).5b Thus, both 1A2B and its
Recognition of the Effector and Recall of the Information.
To show that the system had been trained for recognition of
the Zn(II) cation effector, zinc triflate (5 mM) was added again
to the mixture above, resulting in the immediate regeneration of
the Zn(1A2B)2 complex with the preformed ligand constituent
in the unchanged distribution obtained after removal of the
cations (see above). The response of the system to this new
addition of zinc cations took less than the time needed to
2
1
agonist A1B were up-regulated, whereas the amounts of A1B
1
measure the H NMR spectrum (about 2 min) and to observe
2
and A2B decreased substantially. These changes amount to a
the formation of the complex Zn(1A2B)2. This fast recall on
second contact with zinc cations as compared to the first
exposition results from the fact that the system does not need
to modify its constitution, as it is optimized for binding zinc
cations. It is probable that the time to form the zinc complex
was less than a few seconds on this second exposition as
compared to several hours for the first exposition under the
same conditions. Thus, the renewed presentation of the effector
zinc(II) to the system induced a fast recall of the encoded
information, and showed that the system had been trained for
recognition of the effector (see step 4 in Scheme 1).
storage of information in the constitutional distribution
triggered by addition of zinc triflate (Zn(OTf)2). The reaction
time of these metallo-induced changes increased markedly
when the temperature was decreased to about 12 h at 60 °C.
Starting from this distribution of 1A1B, Zn(1A2B)2, 2A1B, and
2A2B obtained by metalloselection, removal of the zinc(II)
cation from Zn(1A2B)2 was achieved by addition of potassium
ferricyanide (5 mM, 1 equiv with respect to Zn(II) ions), which
resulted in precipitation of both the Zn(II) and the Sc(III)
cations, thus removing all cations from the medium. Because
the exchange between the hydrazones under these conditions is
very slow, the DCL retained its composition unchanged (see
step 3 in Scheme 1, Figure S28) and thus kept the encoded
information in the form of the frozen distribution of
constituents, which did not change at 25 °C for more than 7
days. It is important to note that the system is now in a
Erasing of the Information. The information stored into the
composition of the CDL was erased by removal of the effector
followed by heating, which promoted a restoration of the initial
distribution of constituents (step 5, Scheme 1). The effector
was again removed by addition of potassium ferricyanide (1
equiv with respect to Zn(II) ions) to precipitate the cation
C
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
from its Zn(1A2B)2 complex. Subsequent heating of the
solution at 170 °C for 20 min in the microwave reactor in
the presence of scandium triflate (2 mM) as exchange catalyst
provided the same initial distribution of the four hydrazones
1A1B, 1A2B, 2A1B, and 2A2B, thus closing the circle and
demonstrating the reversibility of all processes during
information imprinting, storage, recall, and final erasing.
Double Training of a Dynamic Covalent Library of
Imines and Acylhydrazones toward Two Different
Effectors. Searching for a Dual Responsive System −
Strong Orthogonal Amplification of a Biased DCL. Setting
the Stage. A further step consists of devising a DCL that may
respond to two different effectors, so that, on applying them, it
may be trained to store different information and undergo
switching between two orthogonal memory states (Scheme 2).
and the hydrazide 1C components (Scheme 3). It is expected to
contain the four constituents A1C, A2 E, A1C, and A2 E,
which could in principle respond to dual training by either
Zn(II) and Cu(I) cations. [One should note that the rate of
formation of the acylhydrazones is increased in the presence of
aliphatic or aromatic amines (see Schemes 3 and 4). Indeed,
amines act as nucleophilic catalysts for the formation of
acylhydrazones (see ref 19). This was demonstrated here in the
case of the formation of E-3A1C from its components (3A and
1C), which was substantially faster (from days to hours) and
occurred at room temperature (down from 70 °C) in the
presence of a catalytic amount (∼10%) of p-toluidine (see also
5
5
1
6
6
1
3
E/Z isomerization of A1C in the Supporting Information).]
1
5
6
When all four components C, A, A (1.25 × 10−5 mol, 25
mM), and E (0.625 × 10−5 mol, 12.5 mM) were mixed in a
1
NMR tube in acetonitrile at 23 °C, the only components
identified in the mixture after about 12 h were A1C (E form
5
Scheme 2. Dual Responsive Dynamic Combinatorial Library
System Displaying Amplification of Two Orthogonal Pairs of
Agonists by Component Selection in Response to Two
Different Metal Cations, Thus Undergoing Double Training
only; see Figures S1−S6 for more details on E and Z
formation) and 6A2 E, indicating a strongly biased initial library.
1
3
5
[The E/Z isomerization of A1C and A1C was studied in
1
a
detail. All four isomers were observed and characterized by H
with Network Switching
NMR (1D and 2D). E-3A1C and Z-5A1C were isolated in pure
forms, and their transformation from one isomer to the other
was achieved by various methods. The other two isomers were
converted into their counterpart in solution. Details about the
E/Z isomerization of these compounds are given in the
Supporting Information. For another study on E/Z isomer-
ization of acylhydrazones, see the literature.18] The origin of
this bias may be attributed to the favored formation of the
acylhydrazone 5A1C and to the stabilization of its agonist 6A2 E
1
by hydrogen bonding between the phenolic −OH and the
15
imine −N (see also Figure S29). Both factors combine to
drive the selection between constituents toward a strong
5
6
1
amplification of the agonists A1C and A2 E in a synergistic
fashion even in the absence of any metal cation.
Addition of zinc cations led to partial destruction of the
6
1
library due to extensive hydrolysis of the bis-imine A2 E.
On the other hand, addition of the copper(I) cation caused
5
the strong amplification of the opposite agonist pair A1E and
1
6A1C via formation of the Cu(5A2 E) complex (characterized
by its NMR data and its particularly clean ESI mass spectrum;
see Figures S33−S35). Thus, a remarkable change of
constituent expression had taken place with a very pronounced
difference between the effects of the two different cations and
an orthogonal switching of the DCL constituents with respect
to both the initial DCL and its composition in the presence of
Zn(ll) cations (see also Figures S32 and S36). However, in view
of both the extensive hydrolysis brought about by the addition
of the Zn(II) cations and the initial bias of the DCL, further
exploration toward a suitable system was conducted.
a
The DCL distributions on the bottom left and right represent
orthogonal constitutional imprints, out-of-equilibrium states.
The design of such systems requires a slow component
exchange between the constituents so as to retain the
constituent selection obtained after removal of each of the
two effectors, so as not to lose the information stored before
reading it by a subsequent exposure to each effector. To achieve
such double effector responsive and dual information storing
dynamic system, the DCL must be based on components
capable of generating constituents that respond specifically to
each of the two effectors. To this end, a search for a DCL based
on both imine and acylhydrazone ligand constituents,
presenting respectively bidentate and tridentate binding
subunits, was set up. It was surmised that such a DCL would
be able to respond to metal cations presenting either
tetrahedral or octahedral coordination geometry under metal-
loselection.
H-bonding was eliminated by exchanging salicylaldehyde
(6A) first for benzaldehyde and later to further reduce
hydrolysis, for 4-chlorobenzaldehyde (4A). Replacing the
aliphatic amine by an aromatic amine (p-toluidine) then
reduced the level of metal-induced hydrolysis to one-half of the
previous one. A final adjustment consisted of exchanging 6-
methyl-2-pyrydinecarboxaldehyde (5A) for 6-phenyl-2-pyridi-
necarboxaldehyde (3A), as it might enhance binding of
copper(I), considering the known increase in stability of the
copper(I) complexes from 6,6′-dimethyl-2,2′-bipyridine to 6,6′-
diphenyl-2,2′-bipyridine.16
Searching for a Suitable DCL. We first explored a DCL
The new CDL was then generated from an equimolar
mixture of the four components: 6-phenylpyridine-2-carbox-
5
6
1
generated from the two aldehyde A and A, the diamine E,
D
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
5
5
1
6
6
1
Scheme 3. Distribution of the Four Constituents of the DCL A1C, A2 E, A1C, and A2 E without Any Metal (Middle), after
Zn(II) Addition (Top), and after Cu(I) Addition (Bottom) Showing the Biased Initial Distribution (Middle) and Extensive
a
6
1
Hydrolysis of A2 E (Top)
a
5
(a) The component A1C can exist in E and Z forms (see Figures S1−S6 for details). (b) Error on determination ∼ 2%.
Training of the DCL generated above is achieved by subjecting
it to the action of a given metal cation, thus generating a
specific distribution of the constituents, which encodes/stores
the information about that particular metal cation. Removal of
the metal thereafter conserves the specific distribution, leaving
the cation-free system prepared for subsequent operations of
recall, retraining, or erasing.
Scheme 4. Distribution of the Constituents of the Optimized
DCL Generated from the Components in the Absence of
Metal Cations
a
Operating on Cation 1, Zn(II). Training. When the initial
mixture of four components was subjected to addition of
Zn(OTf)2 (0.5 equiv, 0.625 × 10−5 mol, 12.5 mM), the
octahedral coordination to the cation promoted the formation
of the tridentate ligand constituent 3A1C as its complex
Zn(3A1C)2, and as a consequence the amount of its agonists
4A1D increased as well (Scheme 5). The formation of the Zn
library required heating to 60 °C for 3 h leading to full
amplification of only one agonist pair (one diagonal of the
square network; Scheme 5) with the composition: Zn-
(3A1C)2/3A1D/4A1C/4A1D/hydrolysis = 50/<1/<1/36/12%
(see also Figures S43−S45). For identification purposes, the
zinc complex Zn(3A1C)2, as well as the copper complex
Cu(3A1D)2, below have been separately prepared and
characterized (see Figures S41, S42 and S51, S52).
a
The amount of 3A1C indicated is the sum of the E and Z forms of this
constituent.
aldehyde (3A), 4-chlorobenzaldehyde (4A), benzhydrazide
(1C), and p-toluidine (1D) (1.25 × 10−5 mol each, 25 mM)
in acetonitrile, heated at 60 °C for 12 h to achieve equilibration.
Storage. The removal of the zinc cations turned out to be
particularly challenging because of the presence of the N−H
group in the acylhydrazone moiety and of the Lewis acidity of
the zinc cation.
Four constituents A1C, A1D, A1C, and A1D were obtained
with a distribution of 35/12/16/23%, respectively, together
with 14% hydrolysis (Scheme 4, Figures S37−S40). The
3
3
4
4
The acylhydrazone N−H group becomes easily ionizable
upon binding of the zinc cation, resulting in a complex acido−
basic equilibrium and hard to resolve mixtures upon
decomplexations. Further, demetalation reagents (e.g., hex-
acyclen) may interfere with the subsequent steps (retraining,
recall, erasing), or are insoluble in pure acetonitrile (e.g.,
K3[Fe(CN)6]), or catalyze fast re-equilibration (erasing the
information). After considerable exploration, the most suitable
reagent proved to be the bis-tetraethylammonium salt of
EDTA, (NEt4)2EDTAH2. It forms strong complexes with the
Zn(II) cations, which precipitate from the mixture, thus
elevated level of hydrolysis of A1D is part of the equilibrium
4
and could not be avoided. Moreover, the presence of the two
configurational isomers of A1C (E-3A1C (40%) and Z-3A1C
3
(60%), confirmed by 2D NMR; see Figures S13−S20) was
observed, and both were taken into account. DCLs of the same
composition (within experimental error ∼ 2%) can be
generated starting from a mixture either of the four
components or of pairs of agonist constituents.
Dual Responsive Information Processing within a
CDL: Training, Storage, Recall, Erasing, and Retraining.
E
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Retraining. Finally, the addition of Cu(I) triflate (CuOTf,
0.5 equiv, 0.625 × 10−5 mol, 12.5 mM) to the out-of-
equilibrium state followed by heating at 60 °C for 12 h retrains
the system from the zinc-trained/selected DCL composition
directly into the copper-trained/selected one without the need
to pass through an erasing step (see Scheme 7 and Figure S50).
Operating on Cation 2, Cu(I), Scheme 6. The correspond-
ing steps can also be achieved with the second cation Cu(I).
Treatment of the same starting mixture with CuOTf (0.5 equiv,
0.625 × 10−5 mol, 12.5 mM) followed by equilibration at 60 °C
for 12 h led again to a marked change in composition of the
CDL, driven this time by formation of the tetra-coordinated
complex Cu(3A1D)2 (Figure 1).
Scheme 5. Formation of the Zn(II)-Trained DCL
Distribution by Addition of Zn(OTf)2 to a Mixture of the
3
3
4
4
Four Constituents A1C, A1D, A1C, and A1D (Training;
Left Top to Bottom), Which on Cation Removal Mainly
Retains Its Composition in the Metal-Free DCL
(Information Storage; Bottom Left to Right); Recall by
Addition of Zinc(II) (Bottom Right to Left); and Erasing of
the Metal-Free DCL by Heating (Right Bottom to Top) with
Return to the Thermodynamically Driven Initial State
It thus resulted in the amplification of both the Cu(I) ligand
4
3A1D and its agonist A1C, that is, the constituents linked by
the network diagonal orthogonal to the previous zinc-selected
diagonal. The distribution obtained was 18/31/32/10/9% for
3A1C/Cu(3A1D)2/4A1C/4A1D/hydrolysis, respectively
Scheme 6. Formation of the Cu(I)-Trained DCL
Distribution by Addition of CuOTf to a Mixture of the Four
3
4
3
4
Constituents A1C, A1D, A1D, and A1C (Training; Left
Top to Bottom), Which on Cation Removal Mainly Retains
Its Composition in the Metal-Free DCL (Information
Storage; Bottom Left to Right); Recall by Addition of Cu(I)
(Bottom Right to Left); and Erasing of the Metal-Free DCL
by Heating (Right Bottom to Top) with Return to the
Thermodynamically Driven Initial State
eliminating both the metal ion and the complexing agent from
the medium. However, a slight excess was needed to ensure
that no Zn(II) remained in solution, as traces were found to
catalyze re-equilibration of the metal-free state, thus substan-
tially decreasing the lifetime of the cation-free out-of-
3
equilibrium state. Indeed, the slightly higher levels of A1D
and 4A1C observed after demetalation may be attributed to this
zinc-catalyzed equilibration during the demetalation/precip-
itation step (see Figure S48). When care is taken of this factor,
the out-of-equilibrium state is stable (less than 10% change) for
about 12 h (see NMR spectra in Figure S49).
Recall. After removal of the Zn(II) cations, the system is
now trained and prepared for the repeated zinc recognition,
which is expected to be much faster than the initial
complexation where equilibration of the DCL had first to
take place. Indeed, the readdition of the zinc cations resulted in
direct fast return to the previous complexed state (recall) in a
much shorter time (less than the 2 min required to perform the
NMR measurement) than the initial training step (3 h) (see
also above) (Scheme 5; see Figure S48).
Erasing. This metal-free state of the DCL is a kinetically
trapped out-of-equilibrium state and is stable for 12 h at 23 °C,
due to the sluggishness of component exchange of the
acylhydrazone 3A1C member of the library (see above). Indeed,
harsh conditions are needed for erasing the information and
returning to the thermodynamically stable state, the initial
distribution of constituents, that is, heating the mixture for
several days at 60 °C or applying microwave irradiation (40
min, 160 °C, 270 W) (see Figures S39, S40, and S49).
(Scheme 6 and Figures S53, S54). Even though the constituent
3A1C can possibly bind Cu(I) in the same way as 3A1D, the 1H
NMR does not show any direct binding (no significant shift of
signals) as well as no signals of possible mixed-ligand
complexes.
The Cu(I)-imprinted cation-free DCL can thereafter be
generated by addition of NBu4CN, leading to precipitation of
CuCN from the mixture. The resulting out-of-equilibrium state
F
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Scheme 7. Representation of All of the Transformations Performed by the Dual Responsive Dynamic Covalent System Based on
a
3
4
3
4
the DCL A1C, A1D, A1D, and A1C
a
(1) DCL generation from the components; 12 h, 60 °C, CH3CN; (2,3) adaptation/training of the system on Zn(II) effector; 0.5 equiv of
Zn(OTf)2, 3 h, 60 °C; (4) Zn(II) removal by [(Et4)N]2EDTAH2; retention of the distribution imprint/information storage at rt for 12 h; (5) fast
information recall; re-addition of 0.5 equiv of Zn(OTf)2; (6) information/imprint erase; MW, 40 min, 160 °C; (7,8) adaptation/training of the
system on Cu(I); 0.5 equiv of CuOTf, 12 h, 60 °C; (9) Cu(I) removal by (n-Bu)4NCN; retention of the distribution imprint/information storage at
rt for 12 h; (10) fast information recall; re-addition of 0.5 equiv of Zn(OTf)2; (11) direct retraining on Zn(II); 0.5 equiv of Zn(OTf)2, 3 h, 60 °C;
4
4
(12) direct retraining on Cu(I); 0.5 equiv of CuOTf, 12 h, 60 °C; [3A1C, A1D] (bottom left) and [3A1D, A1C] (bottom right) represent
constitutional imprints, out-of-equilibrium states.
is then stable (less than 10% change) for up to 2 days (see
NMR spectra in Figure S56).
CONCLUSION
■
For both DCLs investigated above, different equilibrium
distributions of constituents characterizing different constitu-
tional states (Zn-DCL and Cu-DCL) are reached, along
different training operations, depending on the applied stimulus
(metal cation). The system exhibits long re-equilibration time
at room temperature, allowing it to retain, in the related metal-
free states, the information stored by training on application of
the metal. These out-of-equilibrium states allow for fast
recomplexation/detection of the metal effector, as the DCL is
already in the preprepared state, demonstrating fast recall. On
introduction of another effector/metal cation, the distribution
of the constituents adapts for the recognition of this other
effector. Finally, subjecting an out-of-equilibrium informed
distribution to equilibrating conditions leads to the restoration
of the initial state of thermodynamic equilibrium, thus erasing
the information stored in the DCL on any effector previously
applied. The present results thus display a full circle training/
The reverse process on readdition of Cu(I) is immediate
(time between mixing and measuring the NMR spectrum; less
than 2 min as compared to 12 h for the initial training step).
(See Figure S55.) Thermodynamic equilibration and erasing of
the Cu(I) imprint can be achieved by microwave irradiation (40
min, 160 °C, 270 W, +2% Sc(OTf)3). (See Figures S40 and
S56.) Finally, addition of zinc triflate to an out-of-equilibrium
Cu-free Cu-DCL results in the direct formation of the Zn-DCL
with complete retraining (change of the distribution) and
switching of the CDN (see Figures S57 and S58).
Thus, the full circle of transformations involved in dual
training and information processing based on the same DCL
has been achieved, demonstrating the reversibility of all paths
and the interconversion of all states of the full network as
summarized in Scheme 7.
G
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
(3) (a) Lehn, J.-M. Proc. Natl. Acad. Sci. U. S. A. 2002, 99, 4763−
4768. (b) Lehn, J.-M. Chem. Soc. Rev. 2007, 36, 151−160. (c) Lehn, J.-
M. Top. Curr. Chem. 2011, 322, 1−32. (d) Lehn, J.-M. Angew. Chem.,
Int. Ed. 2013, 52, 2836−2850.
erasing process based on the generation of constitutional
dynamic engrams7 and operating via CDNs.
It is worth stressing that, whereas usually DCC searches for
fast component exchange and constituent equilibration in a
DCL, here a slow process is required to retain the trained,
informed out-of-equilibrium state, highlighting the virtues of
slowness (which may also apply to other cases)!
One may surmise that a more complex DCL based on more
components of various structural types would generate a great
variety of interconvertible constituents, which, when subjected
to given effectors, would produce different distribution patterns
within the higher order dynamic network underlying the DCL,
thus yielding a constitutional fingerprint/engram, a specific
distribution of constituents for each effector. Such multi-
component and multiresponsive constitutional dynamic net-
works provide avenues toward chemical systems6 of increasing
complexity.17
(4) For a selection of reviews specifically on dynamic covalent
chemistry, see, for instance: (a) Lehn, J.-M. Chem. - Eur. J. 1999, 5,
2455−2463. (b) Rowan, S. J.; Cantrill, S. J.; Cousins, G. R.; Sanders, J.
K. M.; Stoddart, J. F. Angew. Chem., Int. Ed. 2002, 41, 898−952.
(c) Corbett, P. T.; Leclaire, J.; Vial, L.; West, K. R.; Wietor, J.-L.;
Sanders, J. K. M.; Otto, S. Chem. Rev. 2006, 106, 3652−3711.
(d) Ladame, S. Org. Biomol. Chem. 2008, 6, 219−226. (e) Miller, B. L.
Dynamic Combinatorial Chemistry; Wiley: Chichester, 2010. (f) Reek, J.
N. H.; Otto, S. Dynamic Combinatorial Chemistry; Wiley-VCH:
Weinheim, 2010. (g) Hunt, R. A. R.; Otto, S. Chem. Commun. 2011,
47, 847−855. (h) Belowich, M. E.; Stoddart, J. F. Chem. Soc. Rev. 2012,
41, 2003−2024.
(5) (a) Chaur, M. N.; Collado, D.; Lehn, J.-M. Chem. - Eur. J. 2011,
17, 248−258. (b) Vantomme, G.; Jiang, S.; Lehn, J.-M. J. Am. Chem.
Soc. 2014, 136, 9509−9518 and references therein.
(6) For systems chemistry, see, for instance: (a) von Kiedrowski, G.
Angew. Chem., Int. Ed. 2005, 44, 6750−6755. (b) Stankiewicz, J.;
Eckardt, L. H. Angew. Chem., Int. Ed. 2006, 45, 342−344.
(c) Kindermann, M.; Stahl, I.; Reimold, M.; Pankau, W. M.; von
Kiedrowski, G. Angew. Chem., Int. Ed. 2005, 44, 6750. (d) Corbett, P.
T. Angew. Chem., Int. Ed. 2007, 46, 8858−8861. (e) Ludlow, R. F.;
Otto, S. Chem. Soc. Rev. 2008, 37, 101−108. (f) del Amo, V.; Philp, D.
Chem. - Eur. J. 2010, 16, 13304−13318. (g) Ghosh, S.; Mukhopadhyay,
P.; Isaacs, L. J. Syst. Chem. 2010, 1:6, 1−13. (h) Hunt, R. A. R.; Otto,
S. Chem. Commun. 2011, 47, 847−858. (i) Grzybowski, B.; Otto, S.;
Philp, D. Chem. Commun. 2014, 50, 14924−14925. (j) Mattia, E.;
Otto, S. Nat. Nanotechnol. 2015, 10, 111−119. (k) For
implementation of an artificial neural network to chirality determi-
nation, see: Granda, J. M.; Jurczak, J. Chem. - Eur. J. 2014, 20, 12368−
12372. (l) For biological networks, see, for instance: Barkai, N.;
Leibler, S. Nature 1997, 387, 913−917.
ASSOCIATED CONTENT
* Supporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: 10.1021/jacs.6b05785.
■
S
Experimental details, NMR spectra, and synthetic
procedures (PDF)
AUTHOR INFORMATION
Corresponding Author
■
*lehn@unistra.fr
Present Address
†Institute for Complex Molecular Systems and Laboratory of
Macromolecular and Organic Chemistry, Eindhoven University
of Technology, 5600 MB Eindhoven, The Netherlands.
(7) Lehn, J.-M. Angew. Chem., Int. Ed. 2015, 54, 3276−3289.
(8) For imprinting in a distribution of dynamic polymers, see: Fujii,
S.; Lehn, J.-M. Angew. Chem., Int. Ed. 2009, 48, 7635−7638.
(9) For examples of molecular imprinting onto polymers, see, for
Notes
The authors declare no competing financial interest.
instance: (a) Mosbach, K.; Ramstrom, O. Bio/Technology 1996, 14,
̈
163−170. (b) Haupt, K.; Mosbach, K. Chem. Rev. 2000, 100, 2495−
ACKNOWLEDGMENTS
We thank the ERC Advanced grant “SUPRADAPT” for
financial support. J.H. and G.V. thank the Universite
Strasbourg for a doctoral fellowship.
■
2504. (c) Liu, J.-Q.; Wulff, G. J. Am. Chem. Soc. 2004, 126, 7452−
7453. (d) Yan, M., Ramstrom, O., Eds. Molecular Imprinted Materials:
̈
́
de
Science and Technology; Dekker: New York, 2005. (e) Andersson, A.
C.; Andersson, H. S.; Ansell, L. I.; Kirsch, R. J.; Nicholls, N.;
O’Mahony, I. A.; Whitcombe, M. J. J. Mol. Recognit. 2006, 19, 106−
180. (f) Haupt, K. Molecular Imprinting, Top. Curr. Chem., 2012.
(g) Whitcombe, M. J.; Kirsch, N.; Nicholls, I. A. J. Mol. Recognit. 2014,
27, 297−401.
REFERENCES
(1) Lehn, J.-M. Chem. - Eur. J. 2000, 6, 2097−2102.
■
(2) For some reviews on the designed formation of specifically
metallosupramolecular entities, see, for instance: (a) Piguet, C.;
Bernardinelli, G.; Hopfgartner, G. Chem. Rev. 1997, 97, 2005−2062.
(b) Fujita, M. Chem. Soc. Rev. 1998, 27, 417−425. (c) Caulder, D. L.;
Raymond, K. N. Acc. Chem. Res. 1999, 32, 975−982. (d) Saalfrank, R.
W.; Demleitner, B. Transition Metals in Supramolecular Chemistry;
Wiley: New York, 1999; pp 1−51. (e) Leininger, S.; Olenyuk, B.;
Stang, P. J. Chem. Rev. 2000, 100, 853−908. (f) Albrecht, M. J.
Inclusion Phenom. Mol. Recognit. Chem. 2000, 36, 127−151.
(g) Albrecht, M. Chem. Rev. 2001, 101, 3457−3497. (h) Seidal, S.
R.; Stang, P. J. Acc. Chem. Res. 2002, 35, 972−983. (i) Ruben, M.;
Rojo, J.; Romero-Salguero, F. J.; Uppadine, L. H.; Lehn, J.-M. Angew.
Chem., Int. Ed. 2004, 43, 3644−3662. (j) Fujita, M.; Tominaga, M.;
Hori, A.; Therrien, B. Acc. Chem. Res. 2005, 38, 369−378. (k) Steel, P.
J. Acc. Chem. Res. 2005, 38, 243−250. (l) Saalfrank, R. W.; Maid, H.;
Scheurer, A. Angew. Chem., Int. Ed. 2008, 47, 8794−8824.
(m) Smulders, M. M.; Riddell, I. A.; Browne, C.; Nitschke, J. R.
Chem. Soc. Rev. 2013, 42, 1728−1754. (n) For two examples, see:
Funeriu, D. P.; Lehn, J.-M.; Fromm, K. M.; Fenske, D. Chem. - Eur. J.
2000, 6, 2103−2111. (p) Funeriu, D. P.; Rissanen, K.; Lehn, J.-M.
Proc. Natl. Acad. Sci. U. S. A. 2001, 98, 10546−10551.
(10) For the generation of an out-of-equilibrium structural state, see,
for instance: Ratjen, L.; Vantomme, G.; Lehn, J.-M. Chem. - Eur. J.
2015, 27, 10070−10081.
(11) For related examples of conformational dynamics, see, for
instance: (a) Choudhury, A.; Geetha, B.; Sangeetha, N. R.; Kavita, V.;
Susila, V.; Pal, S. J. Coord. Chem. 1999, 48, 87−95. (b) Choudhary, S.;
Morrow, J. R. Angew. Chem., Int. Ed. 2002, 41, 4096−4098.
(c) Bernhardt, P. V.; Chin, P.; Sharpe, P. C.; Richardson, D. R.
Dalton Trans. 2007, 30, 3232−3244. (d) Chow, C. F.; Fujii, S.; Lehn,
J.-M. Angew. Chem., Int. Ed. 2007, 46, 5007−5010. (e) Cao, X. Y.;
Harrowfield, J.; Nitschke, J.; Ramirez, J.; Stadler, A. M.; Kyritsakas-
Gruber, N.; Madalan, A.; Rissanen, K.; Russo, L.; Vaughan, G.; Lehn,
J.-M. Eur. J. Inorg. Chem. 2007, 18, 2944−2965.
(12) For adaptation to shape switching, see: (a) Ulrich, S.; Lehn, J.-
M. J. Am. Chem. Soc. 2009, 131, 5546−5559. (b) Ulrich, S.; Buhler, E.;
Lehn, J.-M. New J. Chem. 2009, 33, 271−292. (c) Ulrich, S.; Lehn, J.-
M. Chem. - Eur. J. 2009, 15, 5640−5645.
(13) For examples of configurational dynamics, see, for instance:
(a) Palla, G.; Mangia, A.; Predieri, G. Annali Chim. 1984, 74, 153−158.
(b) Belov, D. G.; Rogachev, B. G.; Tkachenko, L. I.; Smirnov, V. A.;
H
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Aldoshin, S. M. Russ. Chem. Bull. 2000, 49, 666−668. (c) Syakaev, V.
V.; Podyachev, S. N.; Buzykin, B. I.; Latypov, S. K.; Habicher, W. D.;
Konovalov, A. I. J. Mol. Struct. 2006, 788, 55−62.
(14) Giuseppone, N.; Schmitt, J.-L.; Schwartz, E.; Lehn, J.-M. J. Am.
Chem. Soc. 2005, 127, 5528−5539.
(15) Kovarí
9455.
(16) Dietrich-Buchecker, C. O.; Marnot, P. A.; Sauvage, J. P.;
Kintzinger, J. P.; Maltese, P. Nouv. J. Chim. 1984, 8, 573−582.
̌ ̌
cek, P.; Lehn, J.-M. J. Am. Chem. Soc. 2012, 134, 9446−
̀
(17) For complexity and complex systems, see, for instance: (a)
Reference 3d. (b) Mainzer, K. Thinking in Complexity, 5th ed.;
Springer: Berlin, 2007. (c) Cohen, R.; Havlin, S. Complex Networks;
Cambridge University Press: Cambridge, 2010. (d) Nicolis, G.;
Nicolis, C. Foundations of Complex Systems, 2nd ed.; World Scientific
Publishing Co.: Singapore, 2012. (e) Alfonso, I. Chem. Commun. 2016,
52, 239−250.
̌ ̌
(18) van Dijken, D. J.; Kovarícek, P.; Ihrig, S. P.; Hecht, S. J. Am.
Chem. Soc. 2015, 137, 14982−14991.
(19) (a) Dirksen, A.; Dirksen, S.; Hackeng, T. M.; Dawson, P. E. J.
Am. Chem. Soc. 2006, 128, 15602−15603. (b) Crisalli, P.; Kool, E. T. J.
Org. Chem. 2013, 78, 1184−1189. (c) Cordes, E. H.; Jencks, W. P. J.
Am. Chem. Soc. 1962, 84, 826−831.
I
DOI: 10.1021/jacs.6b05785
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX