Photoswitching of Co(ii)-based coordination cages containing azobenzene backbones

Inclusion of photoswitchable azobenzene units as spacers into ditopic bridging ligands Lm and Lp, containing two chelating pyrazolyl-pyridine termini, allows formation of metal complex assemblies with Co(ii) that undergo a range of light-induced structural transformations. One notable result is the light-induced conversion of a Co2(Lp)3 dinuclear triple helicate (based on the E ligand isomer) to a C3-symmetric Co4(Lp)6 assembly, assumed to be an edge-bridged tetrahedral cage, based on the Z ligand isomer. Another is the preparation of a series of Co4(Lm)6 complexes, of which Co4(E-Lm)6 was crystallographically characterised and consists of a pair of Co2(Lm)2 double helicates connected by an additional two bridging ligands which span the pair of helicate units, giving a cyclic Co4 array in which one and then two bridging ligands alternate around the periphery. A set of Co4(Lm)6 complexes could be prepared containing different ratios of Z : E ligand isomers (0 : 6, 2 : 4, 4 : 2 and 6 : 0) of which Co4(Z-Lm)2(E-Lm)4 was particularly stable and dominated the speciation behaviour, either during light-induced switching of the ligand geometry in pre-formed complexes, or when ligand isomers were combined in different proportions during the preparation. These examples of (i) interconversion between Co2L3 (helicate) and (ii) Co4L6 (cage) assemblies with Lp, and the interconversion between a series of Co4L6 assemblies Co4(Z-Lm)n(E-Lm)6−n with Lm, constitute significant advances in the field of photoswitchable supramolecular assemblies.

Mass Spectrometry -Low resoluPon ESI mass spectrometry was performed using an Agilent 6130B ESI-MS; high resoluPon ESI mass spectra were acquired on a Bruker Compact ESI-Q-TOF.

S24
Thermal Relaxa/on NMR Experiments Ligands Ligands (5 mg, 10 µmol) were dissolved in MeCN-d3 (1 cm 3 ) and irradiated with 340 nm light unPl a photostaPonary state was reached.Spectra were recorded every 30 minutes for 18 hours at 65 °C and every 30 minutes for 60 hours at 25 °C.Z-L p 65 °C

In situ NMR Spectra
We have adapted this method from literature 4 and described it previously. 5We add this descripPon here for convenience: The NMR sample used for the in-situ illuminaPon experiments was fi~ed with an insert tube made from quartz glass.A quartz glass opPcal fibre with a non-terminated end and its exposed surface and the exposed surface roughened to ensure even and omnidirecPonal illuminaPon was subsequently inserted into the insert.In order to avoid damaging the fibre, an aluminium rod was used to lower this construcPon into the NMR spectrometer.The opposite end of the opPcal fibre was connected to the light source to allow for the irradiaPon of the sample inside the spectrometer.

X-ray crystallography
Details of the crystals used, data collecPon and refinements are given in Table 1.The structural determinaPon of ligand L m was performed on a Rigaku Supernova diffractometer.The data was integrated and absorpPon correcPons were applied using a Gaussian numerical method using the CrysAlisPro soiware. 6he data collecPon for E•Co•L m was performed in Experiment Hutch 1 of beamline I-19 at the UK Diamond Light Source synchrotron facility, 7 using methodology, data processing and soiware described previously.1h The structures were solved with Olex2, 8 using dual space iteraPve methods (SHELXT) 9 and refined by a full-matrix least-squares algorithm (SHELXL). 9s is usual with crystallographic structure determinaPons of this kind of elaborate supramolecular assemblies, for E•Co•L m sca~ering is weak and refinement problems are significant due to substanPal disorder, principally of anions and solvent molecules, although the Co4L6 cage superstructure itself showed disorder of some ligand fragments over two closely-spaced posiPons.These problems required (i) extensive use of restraints to ensure geometrically reasonable structures (in parPcular: the geometries of pyrazole and pyridine rings were fixed with AFIX commands due to high disorder, and rigid bond (RIGU) and similarity (SIMU) restraints were applied to the anisotropic displacement parameters of all atoms in the structure; and (ii) eliminaPon of regions of diffuse electron density using the solvent mask feature in OLEX leaving apparent voids in the laƒces equaPng to 513 electrons per complex molecule.Details pertaining to each structure are included in the individual CIFs.Discussion of the structure of E•Co•L m in the main text is accordingly at the level of demonstraPng the gross geometry of the complex with detailed discussion of structural minuPae kept to a minimum.

Structure name
L m E•Co  For E•Co•L p (Figure 7, main text, and S51a), an MM2 force field was needed, because dispersion interacPons in GFN-xTB caused the helicate structure to collapse in order to allow for π-stacking between two of the ligands.Since 1 H NMR suggests approximate D3 symmetry of the E•Co•L p helicate in soluPon, we reverted to a model which does not take dispersion interacPons into account as strongly as GFN-xTB.The models of the verPces were based on the fac vertex obtained from a previously published X-ray crystal structure. 12The cobalt(II) ions were set to octahedral geometry with sp 3 d 2 hybridizaPon to maintain Co-N bond angles.(BF4 − )@E•Co•L p was simulated with a MM2 model for comparison (Figure S51b), showing that the flexibility of the ligands allows them to bend to accommodate the BF4 − guest thereby reducing the Co-Co distance (Figure S51c).The MM2 coordinates for (BF4 − )@E•Co•L p were also subjected to GFN-xTB which resulted in a similar geometry (Figure S51d).S52b) shows that subjecPon of the crystal structure coordinates to the GFN-xTB method causes the structure to unwind significantly, presumably due to the lack of crystal packing constraints.The structure of E•Co•L m seems to exhibit considerable flexibility, so that changing the ligand configuraPon from E to Z does not cause significant changes in the Co4 framework (Figure S52c).

S44
In cyclic bis-azobenzenes, virtually concerted switching of both azobenzenes can be observed, 14,15 with the EZ state being considerably less stable than the EE and ZZ states, respectively, leading to a short lifetime of the EZ state and a long thermal half-life of the ZZ state. 15If the flexibility of the bis-azobenzene macrocycle is not sufficient, switching of the azo-units can be prevented entirely. 16This literature precedence and the fact that we only seem to observe discrete structures with Z/E ratios of 0:6, 2:4, and 6:0, with an apparent fourth species likely possessing a 4:2 ratio, lead us to the assumption that the azo-units of Co•L m would switch in pairs, i.e. the two helicate subunits and the two parallel connecting ligands, reducing the number of possible isomers for the structures with Z/E ligand mixtures, Co•L m •2,4 and Co•L m •4,2, to three, respectively (Figure S53).We optimized the geometries of those possible isomers with GFN-xTB.Structures highlighted in yellow seem more likely to form.

Figure S45 .
Figure S45.Visual presentaTon of the in-situ NMR irradiaTon set up: An NMR sample with a quartz-glass insert and an opTcal fibre.5

Figure S56 .
Figure S56.(a) GFN-xTB structure of E•Co•L m .Comparison of GFN-xTB structure of E•Co•L m (blue) to (b) X-ray crystal structure of E•Co•L m (orange), that subjecTon of the crystal structure coordinates to the GFN-xTB method causes the structure to unwind significantly, and (c) GFN-xTB structure of Z•Co•L m (green), showing that the size of the Co4 metallacycle barely changes upon EàZ isomerisaTon.

Table S1 .
Light output of LEDs used.
340 nm LEDs (only) were unavailable for in situ illuminaPon NMR experiments, but were used for endpoint experiments and synthesis.