Development of an Automated Workflow for Screening the Assembly and Host–Guest Behavior of Metal‐Organic Cages Towards Accelerated Discovery

Abstract

Metal‐organic cages (MOCs) are a class of self‐assembled materials with promising applications in chemical purifications, sensing, and catalysis. Their potential is, however, hampered by challenges in the targeted design of MOCs with desirable properties. MOC discovery is thus often reliant on trial‐and‐error approaches and brute‐force manual screening, which are time‐consuming, costly, and material‐intensive. Translating the synthesis and property screening of MOCs to an automated workflow is therefore attractive, to both accelerate discovery and provide the datasets crucial for data‐led approaches to accelerate MOC discovery and to realize their targeted properties for specific applications. Here, an automated workflow for the streamlined assembly and property screening of MOCs was developed, incorporating automated high‐throughput screening of variables pertinent to MOC synthesis, data curation and automated analysis, and development of a host–guest assay to rapidly assess binding behavior. Computational modelling supplemented this automated experimental workflow for post priori rationalization of experimental outcomes. This study lays the groundwork for future large‐scale MOC screening: from a relatively modest screen of 24 precursor combinations under one set of reaction conditions, 3 clean MOC species were identified, and subsequent screening of their host–guest behavior highlighted trends in binding and the identification of potential applications in molecular separations.

Full text

Research Article
How to cite: Angew. Chem. Int. Ed. 2025, e202424270
doi.org/10.1002/anie.202424270
Metal-Organic Cages
Development of an Automated Workflow for Screening the Assembly
and Host–Guest Behavior of Metal-Organic Cages Towards
Accelerated Discovery
Annabel R. Basford, Aaron H. Bernardino, Paula C. P. Teeuwen, Benjamin D. Egleston,
Joshua Humphreys, Kim. E. Jelfs, Jonathan R. Nitschke, Imogen A. Riddell,*
and Rebecca L. Greenaway*
Abstract: Metal-organic cages (MOCs) are a class of self-assembled materials with promising applications in chemical
purifications, sensing, and catalysis. Their potential is, however, hampered by challenges in the targeted design of MOCs
with desirable properties. MOC discovery is thus often reliant on trial-and-error approaches and brute-force manual
screening, which are time-consuming, costly, and material-intensive. Translating the synthesis and property screening of
MOCs to an automated workflow is therefore attractive, to both accelerate discovery and provide the datasets crucial
for data-led approaches to accelerate MOC discovery and to realize their targeted properties for specific applications.
Here, an automated workflow for the streamlined assembly and property screening of MOCs was developed, incorporating
automated high-throughput screening of variables pertinent to MOC synthesis, data curation and automated analysis,
and development of a host–guest assay to rapidly assess binding behavior. Computational modelling supplemented this
automated experimental workflow for post priori rationalization of experimental outcomes. This study lays the groundwork
for future large-scale MOC screening: from a relatively modest screen of 24 precursor combinations under one set
of reaction conditions, 3 clean MOC species were identified, and subsequent screening of their host–guest behavior
highlighted trends in binding and the identification of potential applications in molecular separations.
Introduction
Metal-organic cages (MOCs) are modular supramolecular
structures that are prepared from both organic and inorganic
building blocks through self-assembly. Metal complexes typi-
cally form the vertices of these polyhedral architectures, while
organic linkers either span the faces or the edges. Their inter-
[*] A. R. Basford, A. H. Bernardino, B. D. Egleston, J. Humphreys,
K. E. Jelfs, R. L. Greenaway
Department of Chemistry, Imperial College London, Molecular
Sciences Research Hub, White City Campus, Wood Lane, London
W12 0BZ, UK
E-mail: r.greenaway@imperial.ac.uk
P. C. P. Teeuwen, J. R. Nitschke
YusufHamiedDepartmentofChemistry,UniversityofCambridge,
Lensfield Road, Cambridge CB2 1EW, UK
I. A. Riddell
Department of Chemistry, University of Manchester, Oxford Road,
Manchester M13 9PL, UK
E-mail: imogen.riddell@manchester.ac.uk
Additional supporting information can be found online in the
Supporting Information section
© 2025 The Author(s). Angewandte Chemie International Edition
published by Wiley-VCH GmbH. This is an open access article under
the terms of the Creative Commons Attribution License, which
permits use, distribution and reproduction in any medium, provided
the original work is properly cited.
nal cavities are capable of binding guests with high affinity
and selectivity, enabling discrimination between chemically
similar molecules and allowing potential applications in chem-
ical purification,[1–3]sensing,[4–7 ]and catalysis.[8–12]Generally,
new MOCs are designed based on concepts uncovered in
previous work, where new structures are created by carrying
out small alterations to the building blocks and tuning the
assembly conditions. The structural diversity of reported
MOCs has been achieved through changing the topicity,
shape, or flexibility of the ligand, the metal ion or counterion
identity, or the solvent, temperature, or concentration.[13,14]
For example, a small change in building block geometry can
lead to a major impact on the resulting MOC architecture
and characteristics.[13,15,16]The exploratory and serendipitous
nature of MOC discovery therefore often leads to wasted
resources when no stable structures are observed under a
range of reaction conditions. The application of both auto-
mated experimental and computational screening, however,
has the potential to help accelerate this design and discovery
process.
Previously, our work has focused on the automated
discovery of organic supramolecular assemblies, including
porous organic cages and catenanes, which was further
streamlined into a low-cost, open-source workflow which
combined automated synthesis, characterization, and
analysis.[17,18]This workflow was designed to be translatable to
other supramolecular architectures. Computational investi-
gations of MOCs to derive design rules and fundamental
Angew. Chem. Int. Ed. 2025, e202424270 (1 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
understanding are still relatively scarce, but notable
contributions from Piskorz et al. and Tarzia et al. demonstrate
the value of using computational screening to better
understand the MOC self-assembly process, enabling desired
architectures to be targeted and experimental outcomes to be
rationalized.[19,20]
To the best of our knowledge, there is only one
recent example demonstrating translation of the assembly
and host–guest screening of MOCs into an automated
mobile workflow,[21 ]and the application of automated high-
throughput screening and analysis methods to the MOC
discovery process has not yet been thoroughly demonstrated.
Additionally, most reported MOC studies only show a subset
of successful reaction parameters, typically not including
those that led to “failures”. Often, the self-assembly results of
reactions that either yielded the same structures, or indeed no
stable structures, are not reported, despite providing valuable
information for future work. The methods explored in this
work will allow for the creation of a more complete picture
of the self-assembly landscape. Not only will an automated
workflow accelerate the experimental screening process, the
creation of MOC self-assembly data in this way will be
important for subsequent use in machine-learning (ML)
methods toward MOC structure and property prediction.
Uptake of ML methods using this type of data will underpin
the future discovery of MOCs with valuable properties, to
provide solutions for key industrial processes governed by
host–guest binding.
Here we utilize and expand on our previously devel-
oped low-cost automated workflow for organic cages and
establish its applicability to discrete metal-organic assemblies
(Figure 1). Automated screening of the self-assembly condi-
tions for a representative family of MOCs is demonstrated,
followed by automated characterization of the reaction
mixtures by using high-resolution electrospray ionization
mass spectrometry and 1H NMR spectroscopy. Analysis of
the reaction mixtures is further aided by automated data
analysis. Computational modelling was also performed to
support experimental outcome rationalization and identify
stability trends. Following subsequent scale-up of a handful
of MOCs, the automated workflow was then expanded to
include an automated host–guest assay for the rapid screening
and identification of promising guest binders.
Results and Discussion
Automated Synthetic Screen
First, a face-capped M4L4tetrahedral (T-symmetric) MOC
was chosen as a representative MOC assembly to study in the
workflow. The Zn4L4MOC selected is known to form via the
simultaneous dynamic covalent imine condensation between
a triamine and monoaldehyde to generate a ligand (L)which
coordinates through the pyridyl and imine nitrogens with
the Zn(II) metal ions (M).[22]This Zn4L4MOC architecture
was selected due to its potential for strong guest binding,[23]
and because derivatives have been shown to encapsulate a
range of hydrocarbons and anions.[24–26]In addition, by simply
changing the metal from Zn(II) to Fe(II), the self-assembly
of a M12L12 pseudo-icosahedral topology was favored, and
when a combination of 1,3,5-tris(4-aminophenyl)benzene (A)
and 2-formyl pyridine (1) was used as the linker components,
both M4L4and M12L12 MOCs were formed depending on
the reaction conditions, alongside helicates (M2L3)atlower
concentrations.[27]By varying the self-assembly precursor
components, our aim was to better understand their influence
on the reaction outcome and demonstrate the potential
of applying an automated workflow to the labor-intensive,
time-consuming task of screening MOC reactions for the
determination of self-assembly reaction outcomes.
Within the automated synthetic screen, the self-assembly
process was conducted at a 4:12:4 equivalence between
the triamine, aldehyde and metal salt building blocks. The
precursor library contained two tritopic triamines (Aand B)
and four aldehydes with varying methylation on the pyridyl
ring (1–4), leading to 8 potential tritopic imine ligands, named
based on their precursor components (L=A1,A2,A3,A4,
B1,B2,B3,andB4), and three metal salts with varying
counterions: Zn(NTf2)2,Zn(OTf)
2,andZn(BF
4)2(Figure 2a).
The reaction stoichiometry was selected to specifically target
aM4L4tetrahedron, however, other self-assembled products
may form including M12L12 icosahedra and helicate structures
formed from a ligand (L) linker and/or an intermediate (I)
ditopic linker (formed by a two-fold partial imine conden-
sation between the triamine and aldehyde precursors): M2I3,
M2I2L,M2IL2,andM2L3(Figure 2b–d).
Automated reaction preparation was carried out on an
Opentrons liquid-handling platform (OT-2),[28]with the deck
containing a 24-well plate of each precursor stock solution, a
6-well solvent plate, and a 48-well reaction plate. A general
OT-2 Python input protocol script was adapted to include
the solvent calibrated gantry, aspirate and dispense speeds,
along with the stock/solvent locations and move commands
dictionaries (full details given in the Supporting Information
S2). The reaction plate was subsequently removed from the
deck, sealed and stirred at 70 °C for 17 h. A 20 µL sample
was then taken, dissolved in MeCN, and taken for HRMS
analysis. The bulk reaction mixture solvent was removed with
a 48-well EquaVAP[29 ]and redissolved in acetonitrile-d3for
1H NMR analysis. Overall, 24 combinations were screened
under the same reaction conditions (overall concentration:
0.0092 M, total volume: 1 mL, MeCN, 70 °C, and 17 h) in
2 mL capped vials with stirrer bars, and repeated to check
reproducibility.
Automated Characterization and Analysis
High-resolution mass spectrometry (HRMS) and high-
throughput 1H NMR spectroscopy were selected as data
characterization techniques for the automated synthetic
screen and used in conjunction with one another. Overall,
HRMS analysis indicated reactivity of the precursors
and identified whether a MNLN(where N=number of
incorporated units) or M2X3(where X=Lor I) species of
interest had formed, while the conversion was analyzed by
Angew. Chem. Int. Ed. 2025, e202424270 (2 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
Figure 1. Summary of the automated workflow for the assembly and property screening of metal-organic cages (MOCs): (Top left, red)
High-throughput automated experimental synthesis, including the setup of an Opentrons OT-2 liquid handing platform for screening the assembly
of a metal ion (M) with two triamines (A and B) and four aldehydes (1–4), yielding eight ligands (L =A1, A2, A3, A4, B1, B2, B3, and B4), screened
with three counterions (C); (Top middle, orange) automated characterization and analysis of 1H NMR spectra and high-resolution mass
spectrometry to assess the assembly outcomes from the synthetic screen; (Top right, purple) outputs of the automated data analysis were used to
identify interesting “hits” to take forward for property screening; (Middle, green) conventional batch scale-up of any identified hits; (Bottom left,
pink) computational modelling for visualization and to help rationalize experimental outcomes; and (Bottom right, blue) automated host-guest assay
of 32 guests to probe binding properties of the scaled up hits from the automated screen.
1H NMR spectroscopy. Where unreacted aldehyde precursor
was observed, this indicated that the reaction had not gone
to completion. This two-step approach filters the results from
the screen, with only full characterization conducted on any
“hits”, although assignment of the 1H NMR spectra was
attempted for all spectra collected to deduce the reaction
outcome and identify the major assembly. To streamline
the analysis of the characterization data, the open-source
code from our cage database tool cagey was adapted to
create both moc_ms_analyser.py and moc_nmr_analyser.py
(https://github.com/GreenawayLab/development-automated
-workflow-mocs), enabling automated analysis.
First, a Python script was developed to calculate likely
possible structures that may form between the six-coordinate
metal ion, tri-topic ligand and the di-topic intermediate
(where the triamine has undergone two imine condensation
reactions with a given aldehyde but a free amine remains),
and with a varying number of counterions. For each reaction,
the metal, triamine, aldehyde, and counterion SMILES were
loaded as a Python dictionary and the empirical formulae
were calculated for the metal ion, tritopic ligand, ditopic
intermediate, and counterion. The combinations of these com-
ponents into possible assemblies, with their corresponding
empirical formulae, charges, and isotope splitting patterns,
were subsequently calculated. The experimental HRMS data
was then searched for the m/z isotope splitting of each
of the calculated possible solutions, and if found, written
to a data frame containing: the formula of the structure,
found m/z, found intensity, predicted m/z, predicted intensity,
and predicted charge. The data frame was then ordered
by formula and the m/z splitting between the peaks was
calculated to assign the charge. The results were filtered down
to formulae found where the observed charge matched the
predicted charge, yielding a machine-readable data frame of
the HRMS results.
Next, the 1H NMR data was analyzed for conversion,
where any residual aldehyde signals indicated that the
dynamic covalent reaction had not gone to full conversion.
Angew. Chem. Int. Ed. 2025, e202424270 (3 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
Figure 2. a) Precursors used within the automated synthetic screen for the assembly of metal-organic cages; b) possible tetrahedral (M4L4), c)
pseudo-icosahedral (M12L12 ), and d) helicate (M2I3,M
2I2L, M2IL2,andM
2L3) architectures, which may form when using this combination of
precursors in a synthetic screen.
Both cage peaks and any residual aldehyde were auto-
matically identified, using the moc_nmr_analyser.py Python
script. The script was based on that from cagey, but further
adapted to include the NMR solvent acetonitrile-d3,asitwas
previously only suitable for chloroform-d. The user benefits
from providing one path to the folder of the NMR raw data
as the input, and then a CSV output labelled by the raw files
title name is written containing the peak type, peak shift in
ppm, and amplitude. If any residual aldehyde was observed,
a further qualitative analysis was performed to determine the
percentage of remaining aldehyde which was categorized and
outputted as a JSON database file. If the relative percentage
of the aldehyde to the largest cage or imine peak was below
1% it was categorized as “minor”, if between 1% and 5%
it was categorized as “still considered for scale-up”, and if
above 5% categorized as ‘conversion not satisfactory’. This
additional step is qualitative as no internal standard was used
but provides additional insight when identifying “hits” to scale
up. Although this level of automated analysis is not suitable
for full characterization and identification of the exact MOC
topological outcome, it does streamline the screening process
by giving an indication of whether a targeted species has
formed and the conversion of the reaction (Figure 3), saving
researcher’s effort and time.
Overall, the reproducibility between the two replicate
screens was in broad agreement – 23/24 (96%) resulted in
the same automated reactivity analysis by HRMS. Taking
into account minor changes in conversion ranging from <1%
to <5%, 20/24 reactions gave reproducible 1H NMR spectra.
Of all the reactions, 75% of precursor combinations contain-
ing triamine Aresulted in the formation of the MNLNor M2X3
Figure 3. Overall results from the automated synthesis screen based on
the automated analysis of HRMS data to identify target structures or
fragments (MNLNand M2X3) as either the major (red) or minor
(orange) assembly, or not identified (grey), and of 1H NMR spectra for
conversion based on residual aldehyde, where aldehyde <1% observed
(red), 1%–5% (orange) and >5% (grey). a) reactions 01–24 results, and
b) repeat reactions 25–48. A fully red box is indicative of the ideal
outcome from the automated analysis, where a reaction has <1%
aldehyde remaining and a target structure has been identified as a major
peak in the HRMS.
Angew. Chem. Int. Ed. 2025, e202424270 (4 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
targeted topologies as the major mass ion peak, whereas none
of the combinations with triamine Bresulted in the targeted
species being identified in the HRMS spectra, suggesting a
larger linker core is required to form a stable self-assembled
species. Of the triamine Bcombinations, a higher proportion
of ligand and intermediates were identified in the HRMS
spectra (Supporting Information S2.3), indicating that while
self-assembly with the metal ion was not successful, the imine
condensations between Band the aldehydes had occurred.
This conclusion was further confirmed by the observation that
92% of the reactions with triamine Bshowed high conversion
based on the automated 1H NMR analysis. This observation
highlights the need for multiple characterization methods
to be used in parallel for interpretation and determination
of the outcome from automated high-throughput screens.
Finally, of the precursor combinations containing triamine A,
a clear effect on the assembly outcome can be observed on
incorporating a methyl group and on varying its positional
isomerism (ortho,meta,orpara to the pyridyl nitrogen): when
the methyl substituent is introduced in the para-position (4),
far away from where metal coordination occurs, the assembly
outcome is similar to the base pyridyl-aldehyde (1), but the
likelihood of successfully forming the targeted MNLNor M2X3
species reduces as the methyl group is moved around the
ring from the para- to meta- (3)toortho-positions (2). We
attributed this observation to increased steric hinderance
around the metal center on co-ordination, which reduces the
likelihood of MOC formation.
Identification and Scale-up of “Hits”
On filtering of the results, six precursor combinations were
identified as “hits” where the automated HRMS analysis
identified the targeted topological mass ions as major species
and automated 1H NMR analysis identified high conversion
with <1% residual aldehyde remaining. However, while the
automated HRMS analysis identifies any targeted MNLNor
M2X3assemblies, the latter M2X3species may also form from
fragmentation of larger topologies during the analysis.[30]
For example, a M2L3mass ion may be from a helicate
or mesocate structure, but may also be a fragment of an
M4L4tetrahedron or higher-order architecture. In addition,
as alluded to above, full consumption of the aldehyde does
not necessarily indicate the formation of a MOC topology
and could instead simply indicate complete conversion to the
tri-imine ligand. Therefore, the 1H NMR spectra of the hits
were manually interpreted further to determine the reaction
outcome in relation to the number of species formed and
subsequently correlated with the identified topologies from
the mass spectra to identify the major assemblies that had
formed (Table 1). Overall, this yielded three MOCs with a
T-symmetric M4L4structure, but with different ligand and
counterion components, (Zn4(A1)4.(NTf2)8,Zn4(A1)4.(BF4)8
and Zn4(A3)4.(BF4)8, referred to as cage 1,cage 2,and
cage 3, respectively (Figure 4). These three cages were
subsequently taken forward and manually scaled-up using the
same reaction conditions for full characterization (Supporting
Information S3).
Table 1 : Hits identified from the automated analysis that were taken
forward for further 1H NMR interpretation to identify whether a single
assembly or a mixture of species had been formed as the major
product(s), and if single, the structural outcome.
Precursor combination Single or Mixed Species? Structural Outcome
A1/Zn(NTf2)2Single M4L4
A1/Zn(BF4)2Single M4L4
A3/Zn(BF4)2Single M4L4
A4/Zn(NTf2)2Mixed –
A4/Zn(OTf)2Mixed –
A4/Zn(BF4)2Mixed –
Figure 4. Scale-up syntheses of the hit M4L4tetrahedron cages: a)
Zn4(A1)4.(NTf2)8(cage 1); b) Zn4(A1)4.(BF4)8(cage 2); and c)
Zn4(A3)4.(BF4)8(cage 3).
In the case of all three scale-ups, the tetrahedra could not
be fully isolated from the minor component of the mixtures,
which was assumed to be a mixture of M2X3helicate and/or
mesocate species as observed by HRMS. By finding the ratio
of the imine N =CHsignal integrations for the tetrahedron
and the M2X3species in each mixture, the molar composition
of the tetrahedron in the mixtures could be calculated – the
signal at 8.5–8.6 ppm corresponds to the 9H resonance of the
helicate (assuming full conversion to M2L3, but could also
include M2I3(6H), M2I2L(7H), and M2IL2(8H)) and the
signal at 8.7–8.8 ppm corresponds to the 12H resonance in the
tetrahedron, similar to the related structures found by Bilbeisi
et al.[23]For these scaled-up samples, this gives approximate
tetrahedron:helicate molar ratios of 1:0.04 in cage 1, 1:0.16 for
Angew. Chem. Int. Ed. 2025, e202424270 (5 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
cage 2, and 1:0.19 for cage 3, but it should be noted that these
ratios were calculated following purification.
Diffusion 1H NMR measurements were acquired on the
isolated samples of cages 1–3 to verify that the assigned peaks
for the M4L4topology correlated with a single assembly in
solution (Section S3.4). Additionally, the measured diffusion
of the secondary peaks present alongside cage 2 and cage
3also indicated these components have a single, smaller
solvodynamic radius than the major signals from the tetra-
hedral cages. In cage 2, the major species has a diffusion
coefficient of 7.05 ×10−10 m2s−1while the minor species
signals gave a coefficient of 8.06 ×10−10 m2s−1, indicating
the solvodynamic radius of the minor species is 1.14 times
smaller than the tetrahedral species. Similarly in cage 3,the
diffusion coefficients indicate that the minor species has a
solvodynamic radius 1.16 times smaller (D =7.60 ×10−10 m2
s−1for the tetrahedral component and D =8.85 ×10−10 m2
s−1for the minor one). In both cases the secondary species
is thus reduced in size by the same amount compared to the
major cage, corresponding to the helicate impurities often
observed in these systems.[23]Finally, in cage 1, the diffusion
coefficient was 1.05 ×10−9m2s−1for the tetrahedral species,
a lower value compared to cages 2 and 3. The discrepancy
between the diffusion coefficients could be due to viscosity
differences between the samples, or due to associated anions
expanding the solvodynamic radii of the cationic assemblies
to differing extents. While the MOC cores are analogous in
structure, their solvodynamic radii cannot thus be compared
directly using the Stokes–Einstein equation.[31]
Effect of Metal Concentration and Precursor Stoichiometry
As previously discussed, structural diversity, including vari-
ations of topicity and shape, can be achieved by changing
a range of factors including the precursor components, such
as the metal or counterion, and their relative stoichio-
metric ratios. Therefore, based on the observed formation
of both tetrahedron and helicate architectures in different
relative ratios in the automated screen (Tables S4–S6),
and subsequent scale-up of cages 1–3 (Zn4(A1)4.(NTf2)8,
Zn4(A1)4.(BF4)8,and Zn4(A3)4.(BF4)8,respectively), an
additional screen was conducted using the developed auto-
mated high-throughput workflow to explore the effect of
varying the concentration of the metal salt, and also the
relative precursor ratio, on targeting the M4L4tetrahedron
versus M2X3helicate. This was undertaken on precursor
combinations of A1 and A3 with counterions NTf2−and
BF4−(Figure 5). Two different triamine:aldehyde:metal pre-
cursor ratios were investigated: 4:12:n (to favor tetrahedron
formation) and 6:12:n (to favor helicate formation), alongside
varying metal salt concentrations from 0% to 100% (n=0–4).
For all combinations across the two stoichiometries inves-
tigated, when no metal salt was present (0%, n=0), minimal
self-assembly and a high proportion of residual aldehyde
was observed, indicating that the metal is required for imine
formation. This is not unexpected–in the presence of water,
the aldehyde-imine equilibrium lies on the aldehyde side,
with the presence of metal favoring imine formation.[32,33]
Figure 5. Overall results from the automated screen exploring the effect
of reaction stoichiometry, ligand, and metal counterion identity and
concentration. Precursor combinations A1/NTf2−(cage 1), A1/BF4−
(cage 2), A3/NTf2−,andA3/BF
4−(cage 3) were investigated at two
reaction stoichiometries, to target the M4L4tetrahedron (4:12:n of
triamine:aldehyde:metal salt) and the M2X3helicate (6:12:n of
triamine:aldehyde:metal counterion), and the metal counterion
concentration was varied between 0% and 100% of 0–4 equivalence
(n=0, 0.8, 1.6, 2.4, 3.2, or 4). Assessment of the remaining aldehyde
following the reaction categorized as either minimal (≤5%, red),
residual (6%–20%, orange), or significant (≥20%, grey). The outcome
of the targeted self-assembly (M4L4or M2X3)aseitherformed(red),a
complex mixture (orange), or minimal (grey). A fully red box is indicative
of the ideal outcome, where a reaction has minimal aldehyde remaining
and a target structure has been identified.
On increasing the metal salt concentration up to 20%−40%
(n=0.8, 1.6), a high proportion of unreacted aldehyde was
still observed alongside a complex mixture of products. As
noted in the initial screen, differences in conversion were
observed between A1 and A3, with the latter showing more
residual aldehyde due to the meta-methyl group causing steric
hindrance around the metal center on coordination, which
reduces the likelihood of MOC formation. A counterion
effect on the degree of assembly was also observed with A3,
with significant aldehyde remaining with NTf2−and greater
conversion observed with BF4−. At increasing (60%–80%)
concentrations of metal salt, while complex mixtures were
typically observed in the 1H NMR spectra, increased starting
material consumption was observed. Although the metal salt
was promoting imine formation, we infer that it was not
present in sufficient quantity to cleanly form the targeted
assemblies. Finally, at the correct stoichiometric ratio of metal
salt (100%, n=4) the formation of the targeted architectures
is observed, and as expected, the four combinations at 4:12:4
triamine:aldehyde:metal salt were in agreement with the
initial screen, where the T-symmetric M4L4tetrahedron was
observed as the major product. However, the stoichiometric
ratio of triamine:aldehyde precursors was found to play an
important role in the outcome of the self-assembly process
–fortheA3/NTf2
−and A3/BF4
−combinations carried out
with a precursor ratio of 6:12:4, clean formation of the singular
M2I3helicate (Zn2(A3-I)3.(BF4)2and Zn2(A3-I)3.(NTf2)2)
Angew. Chem. Int. Ed. 2025, e202424270 (6 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
Figure 6. (Top) GFN2-xTB optimized geometries of eight face-capped
M4L4tetrahedrons assembled from Zn(II) and ligands A1-A4, with
ligand structures shown above. (Bottom) GFN2-xTB optimized
geometries of eight face-capped M4L4tetrahedrons assembled from
Zn(II) and ligands B1–B4, with ligand structures shown above. Facial-
Zn atoms are shown in purple, nitrogen atoms in blue, and carbon
atoms in grey, hydrogen atoms are omitted for clarity.
was observed (Figures S91, S92). This contrasts with the
predominantly clean formation of a tetrahedron from the
same precursors when carried out in a 4:12:4 ratio. In
addition, on manual inspection of the 1HNMRspectrafor
the same stoichiometric precursor ratio with A1/NTf2
−and
A1/BF4
−, the formation of M2X3structures can be assumed
with minimal formation of the M4L4tetrahedron, but one
topology could not be definitively assigned – this complexity
likely arises from the identity of Xleading to a possible mix
of M2I3,M2I2L,M2IL2,andM2L3helicates, but also peak
complexity may arise from the presence of both the meso and
racemic diastereomers,[23,34]limiting comprehensive 1HNMR
assignment and structure confirmation.
Computational Modelling
To better understand the structural outcomes identified in
both experimental screens using the automated analysis,
computational modelling was performed on the MNLN(where
N=4 or 12) and M2X3(where X=Lor I) structures, details
of which are given in Supporting Information Section S5.
First, the M4L4tetrahedra were modelled at a semiempirical
level (Figure 6), with a T-symmetric configuration. For both
triamines Aand B, the MOCs assembled with the methyl-
substituted pyridyl aldehydes 2,3,and4are isomers of each
other, and therefore the relative energies for the different
tetrahedra can therefore be compared (Table 2).
Across both triamines, the same energetic trend is
observed for the M4L4structure (A2 >A3 >A4 and
B2 >B3 >B4). The differences between 3and 4are
relatively small but clearly suggest these are more energet-
ically favorable than structures incorporating aldehyde 2.
This trend broadly agrees with the experimental outcomes
from the automated screen discussed above, in relation to
their reactivity and successful formation of a self-assembled
species. We infer that increased steric hinderance around the
metal center leads to less favorable assembly outcomes due to
their higher relative energies. However, while aldehyde 4in
Table 2 : GFN2-xTB semiempirical energies and r2SCAN-3c//GFN2-xTB
single-point DFT calculations for each of the methylated M4L4tetrahedra
assembled from Zn(II) with ligands A2–A4 and B2–B4. Energies are given
in kJ mol−1relative to tetrahedra with ligands A2 and B2 being set to zero.
Ligand A2 A3 A4
GGFN2-xTB 0−93.7 −102.1
Gr2SCAN−3c//GFN2−xTB 0−85.5 −95.5
Ligand B2 B3 B4
GGFN2-xTB 0−49.3 −58.7
Gr2SCAN−3c//GFN2−xTB 0−43.2 −53.0
the tetrahedral topology was the most energetically favored
across the family, it typically led to the formation of a mixture
of species including a M4L4species (Table 1), and therefore
formation is not solely dependent on the energetics of the
M4L4topology.
Because of the complexities of the M2X3species (where
X=L or I) and the fact that helicate structures may form
based on different chirality at each metal center resulting
in three configurations (homochiral helicates ( and ),
or achiral mesocates ()),[35]further modelling of these
systems was not carried out. In addition, as no pseudo-
icosahedra were experimentally identified, only preliminary
modelling was carried out, which found that ligands formed
with the smaller triamine B, which contains a singular
aromatic ring as the core, were too small to assemble into a
pseudo-icosahedron.
Automated Host–Guest Assay
Prediction of host–guest binding in MOCs is nontrivial to
do reliably via computation, and curation of experimental
data to validate these predictions or feed a data-led approach
remains a bottleneck.[19,20,36]The most common approach
to study and quantify host–guest binding in supramolecular
assemblies is to study the change of one species’ physical
property, such as a chemical resonance in NMR spectroscopy,
or the absorption band in UV–vis spectroscopy.[37 ]Typically
host–guest binding studies using NMR are time consuming
and expensive, requiring a large amount of material and
deuterated solvents, especially for full titrations. Our aim here
was therefore to develop a high-throughput host–guest assay
in a plate-based format for MOCs. The ability to carry out
many single-point NMR binding studies at a known host–
guest equivalence in parallel, performed in an automated way,
allows for the rapid identification of strong binding guests, the
elucidation of trends in binding behavior, and the potential
identification of promising molecular separations.
For this proof-of-concept host–guest assay and to validate
that binding studies can be translated into a high-throughput
automated protocol, the MOC (cage 1,Zn4(A1)4.(NTf2)8),
concentration (0.0008 mmol mL−1), host–guest ratio (1:4),
and range of guests (32 neutral species), were selected based
on literature precedents. For example, cage 1 has previ-
ously been reported to bind neutral guests such as tBuOH
Angew. Chem. Int. Ed. 2025, e202424270 (7 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
and cyclohexane, which were also included in the screen
as controls to validate the workflow.[25,26]The experiment
concentration was selected where equimolar concentrations
of the host and the host–guest complex would give a Kaca. 103
M−1.Thisallowsforstronger(Ka>> 103M−1), intermediate
(Kaࣃ103M−1) and weaker association (Ka<< 103M−1)to
be estimated, based on previously reported ranges of values
for Kain related systems.[25,26]Samples were prepared in
96-well plates using stock solutions on the Opentrons OT-
2 (500 µL of 0.0008 M host in acetonitrile-d3and 75 µL of
0.0214 M guest stock solutions in acetonitrile) before being
sealed and vortexed at room temperature (298 K), and then
transferred to high-throughput NMR tubes on the platform
before being analyzed by 1H NMR spectroscopy after 1 and 7
days (see Supporting Information Section S6 for full details).
The accuracy of automated dispensing was also validated
by checking the obtained host–guest ratio via integration
of free cage 1 and unbound guest (Supporting Information
Figure S102), which confirmed a 1:4 stoichiometric ratio had
been achieved.
From the assay, a host–guest complex was inferred to have
formed if there were two sets of host peaks, one from the
“empty” MOC and the other from the guest⊂MOC in slow
exchange. Observation of changes in the host peak shifts were
selected over changes in the guest peaks due to the aliphatic
nature of many of the guests—these guests were often found
in the upfield region of the 1H NMR spectra and therefore
overlapped with solvent signals, making assignment of free
and complexed guest a less viable approach in our screen. In
cases where encapsulation was observed and slow-exchange
is assumed, the association constant (Ka) was calculated from
the integrations of host peaks relative to the host–guest
peaks, assuming only a singular guest can be encapsulated
(Figure 7, Supporting Information section S6). The calculated
Kavalues from this automated screen give a picture of the
overall binding behavior in a MOC, informing a researcher of
whether encapsulation has occurred and giving an indication
of the relative strength of binding, which in turn can identify
which combinations may be taken forward to perform full
titrations on. Even from a relatively simple screen of 32 guests
with one host, trends may be elucidated and differences in
binding may direct a MOC for use in separations.
For cage 1, the strongest binding guests were carbon tetra-
chloride >cyclohexane >toluene >o-xylene >cyclopentane,
and the proportion of bound guests increased from 25%
(8/32) after 1 day to 69% (22/32) after 7 days, indicating
that guest uptake was slow. Separations of hydrocarbons
from crude oil and alkenes from alkanes have been labelled
as two of the seven chemical separations crucial to future
prosperity.[38 ]In this MOC host-guest assay, o-xylene⊂cage
1was observed but no encapsulation occurred for the p-orm-
isomers; t-butanol⊂cage 1 was also observed but the n-and
s-isomers exhibited no binding, and preferential binding of an
alkane over an alkene was observed with cyclohexane⊂cage 1
versus cyclohexene, with the latter again showing no binding.
After studying the host–guest behavior of cage 1,wethen
turned our attention to comparing the binding of a subset of
guests (toluene, n-octane, carbon tetrachloride, and o-xylene)
in the structurally analogous cages 2 and 3(Figure 8). Cage
Figure 7. A heatmap of the calculated Ka(298 K) values from the
single-point 1H NMR binding assay of the host MOC, cage 1, and 32
neutral guests, for any host–guest complexes formed after 1 day and 7
days. The largest Kavalue between 1 day and 7 days for each host–guest
combination is shown.
Angew. Chem. Int. Ed. 2025, e202424270 (8 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
Figure 8. Comparison heatmaps and the calculated Ka(298 K) values
from the single-point NMR binding assay of the hosts cage 1, cage 2,
and cage 3 with a subset of 4 guests – toluene, n-octane, carbon
tetrachloride, and o-xylene. The largest Kavalue between 1 day and 7
days for each host–guest combination is shown.
1and cage 2 differ only through their counterions, and cage
2and cage 3 only differ based on whether the pyridyl ring is
methylated or not, but computationally, the MOC cores are
similar in size, suggesting cages 1–3 have similar cavity sizes
(Table S7). Therefore, this secondary host–guest screen allows
us to systematically vary the cages by keeping the pyridyl
substitution fixed (cage 1 vs. cage 2) and then the anion fixed
(cage 2 vs. cage 3) to determine the impact of each of these on
any changes to guest recognition and in binding.
Electrostatics can govern host–guest behavior, where a
counterion that is larger than the host void space may
be necessary in order to preserve guest binding capabil-
ity. The bis(trifluoromethylsulfonyl)imide (triflimide, NTf2−)
counterion has been previously chosen for MOCs as a non-
or weakly-coordinating counterion, and in the literature it
was inferred to be too large to fit inside of cage 1.[26]In
contrast, the smaller size of the tetrafluoroborate (BF4−)
counterion would allow possible occupation of the internal
cavity, which may compete for binding with guest molecules.
Nonetheless, the same general trend is observed for cage
1and cage 2 where the binding was observed as follows:
carbon tetrachloride >toluene >o-xylene >n-octane. One
difference between binding was found within toluene⊂cage
2, where binding was slower (not observed after 1 day but
was for 7 days) and weaker by an order of magnitude than
for toluene⊂cage 1. In contrast, the association constant for
carbon tetrachloride⊂cage 2 >carbon tetrachloride⊂cage
1, however, this should be interpreted with care due to
the single-point derivation of these Kavalues. However,
for cage 3, the binding trends differ as follows: carbon
tetrachloride >toluene >n-octane, with no binding of o-
xylene observed. In addition, binding of toluene and n-octane
in cage 3 was slower and only apparent after 7 days, although
n-octane had a higher preliminary Kathan for both cage
2and cage 1. This trend may indicate that methylation
of the tetrahedral MOC may slow guest encapsulation,
along with competitive binding from the tetrafluoroborate
(BF4)−counterion. As binding was slow, a longer time was
required for the host–guest complexes to equilibrate and for
encapsulation to occur.
Overall, both the full guest assay with cage 1,andthe
subset assays with cage 2 and cage 3, have yielded a wealth
of information into the binding behavior of this class of
MOCs. This richness of data is not commonly recorded, nor
is it reported in the literature. This study reveals that even
subtle changes in counterion identity and ligand substitution
have a measurable effect on guest-binding properties, further
contextualizing the requirement for an automated workflow
in both the automated synthesis and property screening of a
range of MOCs on a reasonable timescale to elucidate trends
and find applications within industrial molecular separations.
Conclusion
In conclusion, we have developed an automated low-cost
workflow for rapidly screening the self-assembly landscape
and host–guest behavior of MOCs. The use of an automated
platform for screening the assembly outcomes enables the
parallelization of MOC synthesis, made possible due to the
reduced volume of reactants, specifically ligand, required
per reaction, and the decreased human resource required in
the preparation and analysis of the samples. Automation of
data analysis through a free, simple, open-source workflow
created a method to rapidly assign HRMS and interpret
1H NMR spectra to determine the reaction outcome, which
was based on a combined assessment of the reactivity
by conversion and identification of the topological species,
reducing experimental cost and effort. On rapid filtering
of the results, identification and scale-up of hits was then
carried out. One could argue that a limitation of our
workflow is that analysis was conducted on crude reaction
mixtures without any purification steps, but it should be
noted that self-assembly reactions are often analyzed as
crude mixtures before attempts at isolation are carried out to
identify the equilibrium product, which could be perturbed
on work-up. Indeed, a change in the tetrahedron:helicate
ratio was observed between crude analysis of the automated
screens and after subsequent scale-up and isolation attempts.
However, due to the trade-off between the information
gained and the lack of multiple additional, expensive steps
to achieve purification in an automated manner, this level
Angew. Chem. Int. Ed. 2025, e202424270 (9 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
of analysis was deemed acceptable. Possible extensions of
our automated workflow could include a simple purification
step, such as precipitation of the assembled metal-organic
complexes and filtration through filter plates, enabling anal-
ysis of both the crude and isolated materials, although
handling dispersions can be challenging in an automated
manner. Additionally, the integration of inline or online
analytical techniques could be valuable. For example, using
ReactIR for IR spectroscopy, or automated sampling for
NMR spectroscopic analysis, could enable monitoring of the
assembly process. However, employing these techniques may
be challenging at the elevated temperatures employed in this
study.
Although only a small subset of variables was screened
here under a single set of reaction conditions, the extensive
structural diversity of reported MOCs has been achieved
through variation of parameters that include the metal ion
identity, the metal counterion, topicity of the ligand, the
angle and spacing between the coordinating sites of a ligand,
reaction stoichiometry, solvent choice, overall concentration,
the reaction temperature, and sometimes the inclusion of
a guest molecule to act as a template.[23,27]Specifically, by
tuning the reagent stoichiometries in a subsequent screen
after identification of variable tetrahedron:helicate product
ratios in the initial screen to identify hits, clean formation of
either assembly could be achieved. This workflow could also
be readily extended to include reaction condition screening
and template addition in the automated screening to further
study the influence on the reaction outcome for conversion,
reactivity, and topological outcomes in the future. Pairing
this automated experimental workflow with computational
modelling allowed experimental observations to be rational-
ized post facto. Further work is needed to streamline the
interconnection of these workflows to aid prediction in the
future to direct synthesis.
Finally, our automated workflow was extended to include
screening of host–guest binding. A general knowledge of
binding was readily achieved for a MOC, and trends eluci-
dated, which in the future, may direct further studies and
applications in molecular separations. This was followed
by systematic variation of the MOC, keeping the ligand
functionality and then the anion fixed, to determine the
impact of each of these factors on guest binding. In addition,
while only a single solvent was used for the host–guest assay
here, the workflow could be expanded to include different
solvents, additional guests, a wider variety of MOCs, and
different operational conditions. Furthermore, there is scope
to expand the methods used for screening host–guest binding,
for example, UV–vis measurements, as inline automated
analyses or automated preparation of samples in a plate-based
format for offline analysis, could be used for determining
association constants by titration. The dynamic nature of the
assemblies, along with solvent and counterion effects, can
drastically influence a MOC’s ability to bind a guest. This
limits our current ability to accurately predict binding through
simulation, so screening this property in a high-throughput
approach could be used to generate a host–guest binding
database which can train predictive models for host–guest
pairing with a high degree of accuracy.
Supporting Information
The authors have cited additional references within the
Supporting Information.
Acknowledgements
This work was financially supported by the Engineering and
Physical Sciences Research Council (EPSRC, EP/W01601X/1,
EP/V03457X/1). K.E.J., I.A.R., and R.L.G. thank the Royal
Society for University Research Fellowships and associated
Enhancement Awards. A.R.B. thanks React CDT for funding
(EP/S023232/1). P.C.P.T. acknowledges the Engineering and
Physical Sciences Research Council via project EP/S024220/1
EPSRC Centre for Doctoral Training in Automated Chem-
ical Synthesis Enabled by Digital Molecular Technologies.
stk,stko, and OPTIM calculations were performed using
resources provided by the Cambridge Service for Data
Driven Discovery (CSD3) operated by the University of
Cambridge Research Computing Service (www.csd3.cam.ac.
uk), provided by Dell EMC and Intel using Tier-2 funding
from the Engineering and Physical Sciences Research Council
(capital grant EP/T022159/1), and DiRAC funding from the
Science and Technology Facilities Council (www.dirac.ac.
uk). We acknowledge Callum R. John for developing the
Opentrons OT-2 python script for transfer of samples from
a 96-well plate directly to the high-throughput NMR tubes.
Conflict of Interests
The authors declare no conflict of interest.
Data Availability Statement
The data that support the findings of this study are
openly available in Zenodo at https://doi.org/10.5281/zenodo.
14183035, and the associated code for data curation
and automated analysis is available at https://github.com/
GreenawayLab/development-automated-workflow-mocs.
Keywords: Automation •Host–Guest behavior •Metal-Organic
cages •Self-assembly
[1] Q.-W. Zeng, L. Hu, Y. Niu, D. Wang, Y. Kang, H. Jia, W.-T. Dou,
L. Xu, Chem. Commun. 2024,60, 3469–3483.
[2] X. Zhang, D. Zhang, C. Wei, D. Wang, R. Lavendomme, S. Qi,
Y. Zhu, J. Zhang, Y. Zhang, J. Wang, L. Xu, E.-Q. Gao, W. Yu,
H.-B. Yang, M. He, Nat. Commun. 2024,15, 3766.
[3] C. Zhao, N. Wang, L. Wang, H. Huang, R. Zhang, F. Yang, Y.
Xie, S. Ji, J.-R. Li, Chem. Commun. 2014,50, 13921–13923.
[4] D. Zhang, T. K. Ronson, Y.-Q. Zou, J. R. Nitschke, Nat. Rev.
Chem. 2021,5, 168–182.
[5] B. Paul, R. Natarajan, Inorg. Chem. 2024,63, 8449–8461.
[6] Z. Meng, F. Yang, X. Wang, W.-L. Shan, D. Liu, L. Zhang, G.
Yua n, Inorg. Chem. 2023,62, 1297–1305.
[7] Y. Cui, Z.-M. Chen, X.-F. Jiang, J. Tong, S.-Y. Yu, Dalton Trans.
2017,46, 5801–5805.
Angew. Chem. Int. Ed. 2025, e202424270 (10 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
[8] Y. Xue, X. Hang, J. Ding, B. Li, R. Zhu, H. Pang, Q. Xu, Coord.
Chem. Rev. 2021,430, 213656.
[9] D. M. Kaphan, M. D. Levin, R. G. Bergman, K. N. Raymond, F.
D. Toste , Science 2015,350, 1235–1238.
[10] M. Yoshizawa, M. Tamura, M. Fujita, Science 2006,312, 251–254.
[11] Y. Li, J. He, G. Lu, C. Wang, M. Fu, J. Deng, F. Yang, D. Jiang,
X. Chen, Z. Yu, Y. Liu, C. Yu, Y. Cui, Nat. Commun. 2024,15,
7044.
[12] M. D. Pluth, R. G. Bergman, K. N. Raymond, Science 2007,316,
85–88.
[13] R. Chakrabarty, P. S. Mukherjee, P. J. Stang, Chem. Rev. 2011,
111, 6810–6918.
[14] D. Luo, X.-W. Zhu, X.-P. Zhou, D. Li, Chem. – Eur. J. 2024,30,
e202400020.
[15] Q.-F. Sun, J. Iwasa, D. Oqawa, Y. Ishido, S. Sato, T. Ozeki, Y. Sei,
K. Yamaguchi, M. Fujita, Chem. – Asian J. 2010,11, 216–220.
[16] J. A. Davies, T. K. Ronson, J. R. Nitschke, Chem 2022,8, 1099–
1106.
[17] R. L. Greenaway, V. Santolini, M. J. Bennison, B. M. Alston, C. J.
Pugh, M. A. Little, M. Miklitz, E. G. B. Eden-Rump, R. Clowes,
A. Shakil, H. J. Cuthbertson, H. Armstrong, M. E. Briggs, K. E.
Jelfs, A. I. Cooper, Nat. Commun. 2018,9, 2849.
[18] A. R. Basford, S. K. Bennett, M. Xiao, L. Turcani, J. Allen, K. E.
Jelfs, R. L. Greenaway, Chem. Sci. 2024,15, 6331–6348.
[19] A.Tarzia,K.E.Jelfs,Chem. Commun. 2022,58, 3717–3730.
[20] T. K. Piskorz, V. Martí-Centelles, T. A. Young, P. J. Lusby, F.
Duarte, ACS Catal. 2022,12, 5806–5826.
[21] T. Dai, S. Vijayakrishnan, F. T. Szczypi ´
nski, J.-F. Ayme, E.
Simaei, T. Fellowes, R. Clowes, L. Kotopanov, C. E. Shields, Z.
Zhou, J. W. Ward, A. I. Cooper, Nature 2024,635, 890–897.
[22] Y. Jin, Q. Wang, P. Taynton, W. Zhang, Acc. Chem. Res. 2014,47,
1575–1586.
[23] R. A. Bilbeisi, J. K. Clegg, N. Elgrishi, X. de Hatten, M.
Devillard, B. Breiner, P. Mal, J. R. Nitschke, J. Am. Chem. Soc.
2012,134, 5110–5119.
[24] W.-Q. Xu, Y.-H. Li, H.-P. Wang, J.-J. Jiang, D. Fenske, C.-Y. Su,
Chem. – Asian J. 2016,11, 216–220.
[25] A. M. Castilla, T. K. Ronson, J. R. Nitschke, J. Am. Chem. Soc.
2016,138, 2342–2351.
[26] A. Jiménez, R. A. Bilbeisi, T. K. Ronson, S. Zarra, C. Woodhead,
J. R. Nitschke, Angew. Chem. Int. Ed. 2014,53, 4556–
4560.
[27] D. Zhang, Q. Gan, A. J. Plajer, R. Lavendomme, T. K. Ronson,
Z. Lu, J. D. Jensen, B. W. Laursen, J. R. Nitschke, J. Am. Chem.
Soc. 2022,144, 1106–1112.
[28] “OT-2 Liquid Handler | Opentrons Lab Automation from
$10,000 | Opentrons,” can be found under https://opentrons.com/
products/robots/ot-2/,2023.
[29] “EquaVAP® 48-Well Evaporator (23048),” can be found under
https://www.analytical-sales.com/product/equavap-48-well-
evaporator-23048/,2023.
[30] C. Drechsler, A. Baksi, A. Platzek, M. Acar, J. J. Holstein,
C. J. Stein, G. H. Clever, Chem. Sci. 2024,15, 19264–
19272.
[31] C. C. Miller, J. Walker, Proc. R. Soc. Lond. Ser. Contain. Pap.
Math. Phys. Character 1997,106, 724–749.
[32] D. H. Busch, Record Chem. Prog. 1964,25, 107.
[33] M. M. Blight, N. F. J. Curtis, J. Chem. Soc. 1962,0, 3016.
[34] F. J. Rizzuto, P. Pröhm, A. J. Plajer, J. L. Greenfield, J. R.
Nitschke, J. Am. Chem. Soc. 2019,141, 1707–1715.
[35] Z. Zhang, D. Dolphin, Chem. Commun. 2009, 6931–6933.
[36] A. J. McConnell, Chem. Soc. Rev. 2022,51, 2957–2971.
[37] P. Thordarson, Chem. Soc. Rev. 2011,40, 1305–1323.
[38] D. S. Sholl, R. P. Lively, Nature 2016,532, 435–437.
Manuscript received: December 11, 2024
Revised manuscript received: March 21, 2025
Accepted manuscript online: April 06, 2025
Version of record online: ,
Angew. Chem. Int. Ed. 2025, e202424270 (11 of 11) © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH

Research Article
Research Article
Metal-Organic Cages
A. R. Basford, A. H. Bernardino,
P. C. P. Teeuwen, B. D. Egleston,
J. Humphreys, K. E. Jelfs,
J. R. Nitschke, I. A. Riddell*,
R. L. Greenaway* e202424270
Development of an Automated Workflow
for Screening the Assembly and Host–
Guest Behavior of Metal-Organic Cages
Towards Accelerated Discovery
An automated experimental workflow
to accelerate the discovery of metal-
organic cages (MOCs), which is often
hindered by slow, manual approaches,
has been developed. Automation of
the synthesis, characterization, and
analysis, is supported by computational
modelling to rationalize the experimen-
tal outcomes, and development of a
host–guest assay elucidated trends in
MOC binding behavior.
Angew. Chem. Int. Ed. 2025, e202424270 © 2025 The Author(s). Angewandte Chemie International Edition published by Wiley-VCH GmbH