A Diverse Array of Large Capsules Transform in Response to Stimuli

The allosteric regulation of biomolecules, such as enzymes, enables them to adapt and alter their conformation to fit specific substrates, expressing different functionalities in response to stimuli. Different stimuli can also trigger synthetic coordination cages to change their shape, size, and nuclearity by reconfiguring the dynamic metal–ligand bonds that hold them together. Here we demonstrate an abiological system consisting of different organic subcomponents and ZnII metal ions, which can respond to simple stimuli in complex ways. A ZnII20L12 dodecahedron transforms to give a larger ZnII30L12 icosidodecahedron through subcomponent exchange, as an aldehyde that forms bidentate ligands is displaced in favor of one that forms tridentate ligands together with a penta-amine subcomponent. In the presence of a chiral template guest, the same system that produced the icosidodecahedron instead gives a ZnII15L6 truncated rhombohedral architecture through enantioselective self-assembly. Under specific crystallization conditions, a guest induces a further reconfiguration of either the ZnII30L12 or ZnII15L6 cages to yield an unprecedented ZnII20L8 pseudo-truncated octahedral structure. The transformation network of these cages shows how large synthetic hosts can undergo structural adaptation through the application of chemical stimuli, opening pathways to broader applications.

spectrophotometer with a 1 mm path-length cuvette at 25 °C. Circular Dichroism was performed on an Applied-Photophysics Chirascan CD spectrometer using a 1 mm pathlength cuvette. Experiments were recorded at 298 K, maintained with a Peltier temperature control. Measurements were background subtracted from blank solvent in an identical cuvette. The sample concentrations were adjusted to maintain a HV below 800 V.
Small-angle X-ray scattering. For SAXS measurements, all the SAXS profiles were acquired at the B21 beamline of the Diamond light source in the UK, a dedicated beamline for small angle X-ray scattering. The experiment was performed with the SAXS detector (EigerX 4M, Dectris) positioned 3.7 m from the sample with the beamstop positioned at the top of the detector to give access to the widest possible q (scattering wave vector) range for the data collection. The X-ray energy was 13.0 KeV. The q range was calibrated using silver behenate.
Each supplied stock solution was diluted 2-fold and 10-fold in deuterated acetonitrile to give a range of concentrations for the measurements. Samples were filled into 1.5mm borosilicate capillaries for measurement and sealed using parafilm to prevent evaporation.
Capillaries were presented on the beamline using purpose designed 3D printed capillary holders to prevent the delicate borosilicate glass from breaking during sample changeover.
The SAXS data were reduced using DAWN, the Diamond developed data analysis workbench application (available for download from https://dawnsci.org/), according to standard protocols. The data were reduced, averaged and background subtracted. In some cases, variation in thickness of the capillaries required a careful approach to background correction where the contribution from the solvent background was varied fractionally to account for the different X-ray path lengths. Borosilicate capillaries are hand blown and some natural variation in thickness can occur which should be accounted for, particularly in weakly scattering systems. To avoid over-subtraction of the background, the background scattering from an empty capillary was subtracted from both the scattering from the solvent and samples measurements, thus removing the contribution from the capillary itself. The solvent scattering was then subtracted from the sample scattering, with a multiplication factor X accounting for the variation in X-ray path length from different diameter capillaries. Some of the low q data from the solvent was rejected during data processing (typically due to flares or shadows) and so this also resulted in truncation of the manually corrected dilute sample data.

S5
Modelling of SAXS data from molecular assemblies has been performed using SASView (https://www.sasview.org/) developed by Diamond in collaboration with other central facilities. The analytical solution was prepared at a concentration of 0.55 mg mL -1 for Zn II dodecahedron at 298 K. Acetonitrile was used as blank to conduct background subtraction.

Graph theory analysis of M5nL2n series polyhedral cages
In contrast to the well-known classes of regular convex polyhedra (Platonic, Archimedean, Goldberg polyhedra (smallest dodecahedron 1), graph theory 3 was used to predict the V5nF2n-type cages. Note that the vertices are occupied by metal ions M, and the faces are capped by pentatopic ligands L, therefore V5nF2n = M5nL2n.
Key parameters to describe the V5nF2n-type polyhedra are shown below. Faces (F), Windows (Wm), Edges (E), Vertices (V) and common edges (l) represents the number of pentagonal faces, m-sided faces as windows, edges, vertices, and common edges where pentagons share. The composition V = 5n, F = 2n would be driven by the principle of maximum site whereby all metal ions are coordinatively saturated, and all ligand nitrogen atoms bound to a metal.

I. V5nF2n-type polyhedra with pentagons and 3-sided triangular windows
The constraints of Euler's polyhedron theorem and geometrical analysis require: The solution for above equations can be solved: V = 5n, F = 2n W3 = 4(n-1), l = 6-n Figure S1. M5nL2n series of cages with pentagonal faces and 3-sided triangular windows.

II. V5nF2n type polyhedra with pentagons and 4-sided square/rhombus windows
We extended the graph theory to 4-sided faces as windows The above equations can be solved:

Self-assembly of the cages
Subcomponent A (0.9 mg, 1 µmol), Zn(NTf2)2 (1.1 mg, 1.7 µmol), 2-formyl-6-methyl pyridine C (0.61 mg, 5 µmol) and CD3CN (0.4 mL) were added to a small test tube, and the mixture was sonicated for 1 min and heated at 80 °C overnight. The brownish-yellow  Only one set of broad signals of the ligand arms were observed, presumably due to the similarity of the different ligand arm environments as a result of pyrrole-N random orientations, resulting in signal overlap and broadness.  S9 Figure S5. 13 C NMR spectrum (125 MHz, 298 K, CD3CN) of Zn-1 [Zn20L12](NTf2)40. 13 C peaks are broad presumably due to the many isomers in the system caused by rotationally disordered pyrrole-N and the slow tumbling of the large cage molecules in the solution, thus fewer than the expected number of signals are observed. Figure S6. 19         were added to a NMR tube, and the mixture was sonicated for 1 min and heated at 80 °C overnight. The orange-brown solution was concentrated by blowing with N2 and then Et2O was added. The resulting solid was collected by centrifugation, washed three times with additional Et2O and then vacuum dried to give of truncated rhombohedron G⊂Zn-3 as a brown solid. (br), 7.7 (br), 7.6 (br), 7.5 (br), 7.4 (br), 7.4 (br), 7.3 (br), 7.3 (br), 7.2 (br), 7.1 (br), 6.9 (br), 6.9 (br, 4H), 6.8 (br, 4H), 6.6 (br), 6.5 (br), 6.3 (br, 7H), 6.1 (br), 6.0 (br). 13     S17 G⊂Zn-3. 13 C peaks are broad presumably due to the many isomers in the system caused by rotationally disordered pyrrole-N and the slow tumbling of the large cage molecules in the solution, thus fewer than the expected number of signals are observed.

S22
were added to a NMR tube, and the mixture was sonicated for 1 min and heated at 80 °C overnight. The deep orange-brown solution was concentrated by blowing with N2 and then Et2O was added. The resulting solid was collected by centrifugation, washed three times with additional Et2O and then vacuum dried to give of truncated rhombohedron G⊂Co-3 as a brown solid.      (br), 8.3 -8.2 (br), 7.7 (br), 7.5 (br), 7.5 (br), 6.9 (br), 6.8 (br), 6.6 (br), 6.1 (br), 6.0 (br), 6.0 (br).     13 C peaks are broad presumably due to the many isomers in the system caused by rotationally disordered pyrrole-N and the slow tumbling of the cage molecules in the solution, thus fewer than the expected number of signals are observed. Figure S31. 19

Cage-to-cage transformation from Zn-2 to Zn-3
To a Zn-2 [Zn30L12](NTf2)60 (0.06 µmol) cage solution in CD3CN (0.24 mL) was added G (15 eq., 112 µL, 8 mM, 0.9 µmol), the solution was kept at r.t. in an NMR tube for 40 d to give Zn-3. Heating was also tried to accelerate this process, but only partial conversion was observed. Heating overnight did not result in complete conversion.

X-ray Crystallography
All crystals diffracted to sub-atomic resolution, rendering direct methods unsuitable for structure solution. However, Δ-TRISPHAT (G) anions influenced the stereochemistries of the cages and forced them to crystallize in polar (enantiomorphic) space groups. That enabled the use of anomalous scattering for solving crystal structures. This is a common strategy for crystal structure solution of biomacromolecules and its effectiveness is not dependent on diffraction resolution.
Data for all 5 samples have been collected at different synchrotron beamlines (Table S1 and S2) that enable variable wavelength data collection and absorption edge scans necessary for single-wavelength anomalous dispersion/diffraction (SAD) and multiplewavelength anomalous dispersion (MAD) experiments. Zinc K-edge fluorescence detected X-ray absorption spectra was recorded for Zn-4 samples and cobalt K-edge absorption spectra was recorded for Co-3 sample. Absorption spectra were processed with program CHOOCH 4 to identify appropriate wavelength(s) for data collection. Up to 3 wavelengths were selected for data collection. "Peak" wavelength is the maximum of the anomalous scattering factor f'' curve from CHOOCH. "Inflection" wavelength is the inflection point on the anomalous scattering factor f'' curve, which is also the minimum in anomalous scattering factor f' curve. High energy "remote" wavelength was selected to be around 100 eV higher than the peak wavelength. Each data collection scan was performed after re-centring the crystal to an unexposed part in order to minimise the effects of radiation damage from the previous scan (MAD experiment). If the crystal was too small, then only one scan was collected at the peak wavelength (SAD experiment). MAD data were collected at beamline ID30B of ESRF employing synchrotron radiation (peak=1.28215 Å, inflection=1.28268 Å, remote=1.26514 Å) at 100(2) K and using EIGER2 X 6M 0.45 mm Si sensor detector. 3600 oscillation images were taken around the omega axis for each scan with a 0.2° step and 20 ms exposition time for a total of 720°. The X-ray beam was focused to 50 x 50 μm and attenuated to around 20% S37 transmission. Data integration and reduction were undertaken with the autoPROC 5 , which relies on STARANISO 6 , XDS 7 , AIMLESS 8 and other programs from CCP4 suite 9 software packages.
Note that the same structure with I422 space group was also obtained from truncated rhombohedron Zn-3 sample, confirming truncated rhombohedron Zn-3 inevitably led to its transformation to truncated octahedron Zn-4.
SHELXC 10 was used to estimate substructure (zinc atoms) structure factor amplitudes (anomalous differences) from 3 MAD datasets. They were used by SHELXD 11,12 to identify the position of zinc atoms. SHELXE 13 was used to apply density modification to establish the correct handedness, improve zinc atom positions and produce phases for electron density map generation. Coot 14 was used to visualise and manipulate models and electron density. Molecular modelling software Maestro 15 was used to generate the initial model for the pentakis(tridentate) ligand which was then used to create SHELX and Refmac 16,17 dictionaries with GRADE program 18 . Refmac dictionary was imported to Coot to enable editing of ligand conformation.
The quality of the experimental electron density map produced by SHELXE 19 was sufficient to unambiguously place the ligand using Coot. Remote wavelength dataset was selected for refinement to avoid variation of dispersion coefficients near absorption edge.
SHELX dictionary containing the full set of bond distance and angle restraints (DFIX, DANG, FLAT) was used for structure refinement with SHELXL 20 . Thermal parameter restraints (SIMU, RIGU) were applied to all atoms to facilitate a stable anisotropic refinement. Even with these restraints some thermal parameters remain larger than ideal as a consequence of the high level of thermal motion throughout the structure and especially around the vertex incorporating Zn II and Δ-TRISPHAT.
The nitrogen atoms of the central pyrrole rings could not be resolved and each pentatopic ligand was assumed to be 5-fold rotationally disordered. Therefore, all atoms within the central pentagonal rings of the ligands were modelled as 80% carbon and 20% nitrogen.
Carbon-bound hydrogen atoms were included in idealised positions and refined using a riding model. Disorder was modelled using standard crystallographic methods including constraints and restraints where necessary.
Rounds of restrained structure refinement with SHELXL were alternated with inspection of 2Fo-Fc and Fo-Fc electron density maps in Coot and fitting of additional anion (Δ-TRISPHAT and perchlorate) and solvent (toluene, water) molecules. Dictionaries for all S38 of them were created in the same way as for the ligand. The cage shows a high degree of crystallographic symmetry with 1/8 of the truncated octahedron (one crystallographically unique organic ligand, 1.5 Δ-TRISPHAT, and 2.5 Zn II atoms) in the asymmetric unit. The absolute configuration of the crystal was determined by a density modification procedure in SHELXE and corroborated by the chirality of Δ-TRISPHAT used.
Additional anions per Zn20L8 assembly were required for charge balance. These anions (included as perchlorate in the formula above) were significantly disordered and despite numerous attempts at modelling, including with rigid bodies, no satisfactory model for the electron-density associated with them could be found. Therefore, the SQUEEZE 21 function of PLATON 22 was employed to remove the contribution of the electron density associated with the remaining anions and further highly disordered solvent, which gave a potential solvent accessible void of 60075 Å 3 per unit cell (a total of approximately 11083 electrons). Diffuse solvent molecules could not be assigned to acetonitrile or toluene and were therefore not included in the formula. Consequently, the molecular weight and density given above are underestimated.
CheckCIF of Zn-4 (I422) gives 5 A and some B level alerts. These alerts result from the limited resolution of the data (low value of sine(theta_max)/wavelength, low data to parameter ratio and low bond precision) and the high level of thermal motion around the Zn II vertex and Δ-TRISPHAT.
Data integration and reduction were undertaken with the autoPROC, the same as for Zn-4 (I422). This dataset diffracted to 1.55 Å.
Structure solution and restrained refinement was done with the same software and in the same way as for Zn-4 (I422), except that SHELXC prepared data from only one dataset (SAD experiment). The quality of the experimental electron density map produced by S39 SHELXE with only peak dataset was sufficient to unambiguously place the ligand molecules using Coot.
Despite these measures and the use of synchrotron radiation few reflections at greater than 1.55 Å resolution were observed. Nevertheless, the quality of the data is sufficient to establish the connectivity of the structure. The cage shows a high degree of crystallographic symmetry with 1/4 of the truncated octahedron (two crystallographically unique organic ligands, 2.5 Δ-TRISPHAT, and 5 Zn II atoms) in the asymmetric unit. The absolute configuration of the crystal was determined by a density modification procedure in SHELXE and corroborated by the chirality of Δ-TRISPHAT used.
The anions within the structure also show evidence of disorder and one of the located perrhenate anions were modelled as disordered over two locations. The disordered perrhenate anions were restrained to be approximately tetrahedral and most low occupancy oxygen atoms were modelled with isotropic thermal parameters.
Additional anions per Zn20L8 assembly are required for charge balance. These anions (included as perrhenate in the formula above) were significantly disordered and despite numerous attempts at modelling, including with rigid bodies no satisfactory model for the electron-density associated with them could be found. Therefore, the SQUEEZE 21 function of PLATON 22 was was employed to remove the contribution of the electron density associated with the remaining anions and further highly disordered solvent, which gave a potential solvent accessible void of 61414 Å 3 per unit cell (a total of approximately 14245 electrons). Diffuse solvent molecules could not be assigned to acetonitrile or benzene and were therefore not included in the formula. Consequently, the molecular weight and density given above are underestimated.
CheckCIF of Zn-4 (I222) gives 3 A and some B level alerts. These alerts result from the limited resolution of the data (low value of sine(theta_max)/wavelength, low data to parameter ratio and low bond precision) and the high level of thermal motion around the Zn II vertex and Δ-TRISPHAT.
Structure solution and restrained refinement was done with the same software and in the same way as for Zn-4 (I422), except that SHELXC prepared data from only one dataset (SAD experiment). The quality of the experimental electron density map produced by SHELXE with only peak dataset was sufficient to unambiguously place the ligand molecules using Coot.
The cage shows a high degree of crystallographic symmetry with 1/4 of the truncated octahedron (two crystallographically unique organic ligands, 3 Δ-TRISPHAT, and 5 Zn II atoms) in the asymmetric unit. The absolute configuration of the crystal was determined by a density modification procedure in SHELXE and corroborated by the chirality of Δ-TRISPHAT used.
CheckCIF of Zn-4 (F222) gives 7 A and some B level alerts. These alerts result from the limited resolution of the data (low value of sine(theta_max)/wavelength, low data to parameter ratio and low bond precision) and the high level of thermal motion around the Zn II vertex and Δ-TRISPHAT.

Zn-4 (P4322):
Crystals with composition [Zn20L8]·8(C18Cl12O6P)·7PF6·2(CH2Cl2) [+solvent] and P4322 space group were grown by the diffusion of diethyl ether/DCM to an acetonitrile solution of [Zn30L12]·60(NTf2) containing excess tetrabutylammonium hexafluorophosphate (TBAPF6). SAD data were collected at beamline X06SA of Swiss Light Source employing synchrotron radiation (peak=1.28256 Å) at 100(2) K and using EIGER 16M X detector. 3600 oscillation images were taken around omega axis with 0.2° step and 20 ms exposition time for a total of 720°. X-ray beam was focused to 50 x 50 μm and attenuated to 10% transmission. Data integration and reduction were undertaken with the autoPROC, the same as for Zn-4 (I422). This dataset diffracted to 1.46 Å. Structure solution and restrained refinement was done with the same software and in the same way as for Zn-4 (I422), except that SHELXC prepared data from only one dataset (SAD experiment). The quality of the experimental electron density map produced by S41 SHELXE with only peak dataset was sufficient to unambiguously place the ligand molecules using Coot.
The cage shows a twofold crystallographic symmetry with 1/2 of the truncated octahedron (four crystallographically unique organic ligands, 4 Δ-TRISPHAT, and 10 Zn II atoms) in the asymmetric unit. The absolute configuration of the crystal was determined by a density modification procedure in SHELXE and corroborated by the chirality of Δ-TRISPHAT used.
CheckCIF of Zn-4 (P4322) gives 8 A and some B level alerts. These alerts result from the limited resolution of the data (low value of sine(theta_max)/wavelength, low data to parameter ratio and low bond precision) and the high level of thermal motion around the Zn II vertex and Δ-TRISPHAT. Intriguingly, replacement of toluene with benzene also yielded single crystals, which crystallized in chiral I222 space group with one fourth of ligand in the asymmetric unit, leading to a distorted pseudo-truncated octahedron with approximate D2 symmetry ( Figure S44C, G), with two rhombus windows instead of square windows on the top and bottom compared to Zn-4 with approximate D4 symmetry. Two other different truncated octahedral structures with F222 and P4322 space groups were also obtained by using THF and diethyl ether/CH2Cl2 as anti-solvents, respectively. Comparison of four of the Zn II metal frameworks were shown to demonstrate their different symmetries ( Figure S44).
F222 structure also has D2 symmetry, but the C2 axis at the equatorial position is different compared to the I222 structure, for I222 structure, it goes through the two opposite equatorial windows, in contrast, for F222 one, it goes through the two opposite Zn centres.   PHASER/PARROT maps also enabled fitting of 10 Δ-TRISPHAT molecules attached to cage molecule A, thus confirming the absolute configuration of the cage determined by SHELXE. Weak electron density was identified for possible 11 th Δ-TRISPHAT molecule, but it was close to two-fold crystallographic axis and obviously disordered. Therefore, we couldn't reliably fit it. Peak wavelength dataset was used for the refinement since it has slightly higher resolution than other two and it was also needed for Phaser runs.
Additional DFIX commands were added in refinement to maintain ligand conformations without too short intramolecular contacts. We didn't treat diffuse solvent with SQUEEZE for this structure since cage molecule B has been omitted from the refinement.
Co-3 crystallized in the chiral P3121 space group, with one and half whole cages in the asymmetric unit. Three out of eleven G anions bound at the pores where pentagonal faces share edges. The driving forces were inferred to be arene stacking between phenylene rings from different ligand arms and the tetrachlorocatecholate rings of G, as well as electrostatic interactions between anionic G and cationic Co II centers. All Co II centers were bound by two tridentate ligand arms, with three out of the fifteen Co II centers adopting opposite L handedness, resulting in different Co II ···Co II distances.