Diastereoselective Self‐Assembly of Low‐Symmetry Pd n L2n Nanocages through Coordination‐Sphere Engineering

Abstract Metal‐organic cages (MOCs) are popular host architectures assembled from ligands and metal ions/nodes. Assembling structurally complex, low‐symmetry MOCs with anisotropic cavities can be limited by the formation of statistical isomer libraries. We set out to investigate the use of primary coordination‐sphere engineering (CSE) to bias isomer selectivity within homo‐ and heteroleptic Pd n L2n cages. Unexpected differences in selectivities between alternative donor groups led us to recognise the significant impact of the second coordination sphere on isomer stabilities. From this, molecular‐level insight into the origins of selectivity between cis and trans diastereoisomers was gained, highlighting the importance of both host–guest and host‐solvent interactions, in addition to ligand design. This detailed understanding allows precision engineering of low‐symmetry MOC assemblies without wholesale redesign of the ligand framework, and fundamentally provides a theoretical scaffold for the development of stimuli‐responsive, shape‐shifting MOCs.


Introduction
Metal-organic cages (MOCs) are discrete, porous supramolecular architectures assembled from metal ions/nodes and coordinating ligands. [1]The ability to encapsulate guest molecules within the cavities of MOCs has led to investigations for their use in catalysis, [2] sensing, [3] drug delivery [4] and stabilising reactive species. [5]etailed principles behind the self-assembly of highsymmetry MOCs have been elucidated over the last four decades.To generate more sophisticated systems [6] with advanced functionality, [7] attention has recently turned to the development of methodologies to access lower symmetry cages. [8]These include the design of mixed-ligand (heteroleptic) [9] (Figure 1a) and mixed-metal (heteronuclear) MOCs (Figure 1b), [10] as well as those assembled from lowsymmetry ligands (Figure 1c). [11]Using these approaches, lowsymmetry MOCs have been realised that exhibit shape- [12] and orientation-selective [13] guest binding.
The inherent directionality of unsymmetrical ligands gives rise to multiple possible constitutional isomers of their metalorganic assemblies (Figure 1c).Various strategies have been investigated towards the high-fidelity, isomer-selective selfassembly of low-symmetry ligands. [11]Aside from developing ligands with mixed-denticity donors, [14] these include geometric design parameters, [15,16] use of non-covalent interactions within the ligand backbone, [17] and coordination-sphere engineering (CSE; also known as side-chain directing). [18]CSE strategies can be subdivided into two further categories: those that use attractive interactions, such as hydrogen-bonding, [19] and those that use repulsive interactions, such as steric hindrance.
[23][24] We have previously reported preliminary findings of two systems that use steric parameters, by themselves and in combination with geometric designs, [15] whilst Crowley and coworkers have used hydrogen-bonding interactions to direct formation of a cis-Pd 2 L 4 cage. [19]Aside from these individual examples, CSE strategies to direct the isomer-selective assembly of homoleptic MOCs remains a significant, and under-investigated, challenge.
It was envisaged that combinations of unsubstituted and sterically bulky coordinating groups in unsymmetric ditopic ligands would bias self-assembly with Pd(II) ions towards specific isomers of M n L 2n cages.Given their prior utility in directing the self-assembly of heteroleptic MOCs, picoline [22] (L P ) and quinoline [24] (L Q ) coordinating groups were chosen for investigation.
In this work, the successful use of CSE in the selective synthesis of Pd n L 2n MOC isomers is reported.For both quinoline and picoline ligands, biasing towards assemblies with a 2 : 2 stoichiometry of donors at the metal nodes was observed.Intriguingly, the different coordinating groups were selective for alternative donor arrangements around the metal ions, [25] namely cis (L Q ) and trans (L P ).
Through careful investigation, the diastereoselectivities between cis and trans cages were rationalised, and molecular origins for this effect identified, demonstrating the importance of considering the combined effects of both first and second coordination sphere interactions in the design of these supramolecular systems.This detailed understanding has ramifications for the future design of MOCs, particularly those of low-symmetry, and also stimuli-responsive systems using CSE approaches.
CSE strategies allow the targeted assembly of MOCs with different symmetries whilst maintaining the structural formulation resulting from the design of the core ligand scaffold.As such, this nuanced approach provides a route for precision engineering the shape of low-symmetry MOCs, and the cavity spaces within, towards the development of more sophisticated, functional supramolecular hosts.

Pd 2 L 4 cages
Based on a dipyridyl ligand motif originally reported by Chand and co-workers, [26] L1 Q and L1 P (Figure 2a) were synthesised by ester condensation between commercially available 3-(hydroxymethyl)pyridine and the appropriate carboxylic acid.Each ligand was then combined with Pd(NO 3 ) 2 • 2H 2 O in a 2 : 1 ratio in d 6 -DMSO ([L1] = 40 mM) and heated at 50 °C for 24 h; no further changes were observed by 1 H NMR with prolonged heating.
Analysis by electrospray ionisation mass spectrometry (ESI-MS) indicated formation of Pd 2 L 4 assemblies (C1) for both systems (Figure S22-25 and S53-56).Diffusion-orientated spectroscopy (DOSY) further supported this, with each system displaying diffusion coefficients (D = 9.51×10 À 11 and 11.0×10 À 11 m s s À 1 for C1 Q and C1 P , respectively) consistent with related systems. [26]n both cases a mixture of isomers of C1 was observed to form by 1 H NMR (Figure 2b and 2c), but with a major component arising at a compositional fraction greater than expected from a statistical library (i.e. 25 %), demonstrating successful induction of isomer selectivity.The percentage composition of the predominant species was estimated through a comparison of integrals between methylene signals (H e ) and isolated signals in the aromatic region of the 1 H NMR spectra, giving values of approximately 70 % and 50 % for C1 Q and C1 P , respectively (Figure S21 and S52).
For both major species of C1, nuclear Overhauser effect spectroscopy (NOESY) revealed through-space interactions between signals assigned to the different coordinating groups of L1 (Figure S17 and S50), identifying these as either the cis or trans isomers (Figure 1c).In the case of C1 Q , apparent diastereotopic splitting of the CH 2 signal (J � 14 Hz; Figure 2b) was consistent with formation of cis-C1 Q as the major species.The absence of diastereotopic splitting for the major isomer of C1 P led to the conclusion that this was most likely trans-C1 P .
Related Pd(II) cages are known to encapsulate an NO 3 À anion that can be exchanged for stronger binding halide anions. [27]Upon addition of 1 eq. of Bu 4 NCl to C1 Q and C1 P , encapsulation of Cl À was evidenced by notable downfield shifts of signals assigned to the endohedral protons of the cage (e.g.H a and H h Δδ = 0.49 and 0.65, respectively, for C1 P ; Figure S30 and S60).C1�Cl À were subsequently prepared and isolated through self-assembly of L1 with [Pd(CH 3 CN) 4 ](BF 4 ) 2 in the presence of 1 eq. of Bu 4 NCl.Unexpectedly, although the switch to BF 4 À counterions and encapsulation of a Cl À guest in place of NO 3 À anions did not change the identity of the major host isomer, the selectivity values were altered: [28] for C1 Q the cis isomer fell to � 40 % of the mixture (Figure S32), whilst the trans isomer of C1 P increased to � 70 % (Figure S68).

Molecular origins of diastereoselectivity
Interestingly, the relative energies of the cis and trans isomers of both C1 Q and C1 P , computed using density functional theory (DFT) calculations (HSE06 functional and def2-SVP basis set with implicit DMSO solvation), suggested the trans isomer should be most stable for both systems (ESI section S3).This implied that additional external influences, beyond inherent structural factors, were responsible for the observed speciation.Thus, the question arose: why did C1 Q exhibit selectivity towards the cis isomer?
The SCXRD structures showed exohedral BF 4 À anions located in proximity to the Pd(II) ions [30] for both assemblies (Figure 2f and 2g).For C1 P the steric bulk of the methyl groups resulted in a greater F•••Pd distance compared to C1 Q .Indeed, for C1 Q , interactions between the counterions and CÀ H of both pyridine and quinoline donors were observed 2f).This initially led us to consider that the different diastereoselectivities resulted from differences in interactions between cage and counteranions.Specifically, non-covalent interactions between the BF 4 À anions and C1 Q stabilised the cis isomer to such an extent as to make it lower in energy than the more sterically favourable trans isomer.This hypothesis was probed on two fronts: through dilution ([C1] 10-1 mM) in d 6 -DMSO (Figure 3b, 3c, S75 and S76) to investigate the effect of reducing the anion concentration, and the synthesis of C1 Q �NO 3 in the presence of excess BF 4 À or À OTf to monitor the effect of increased anion concentration and stoichiometry (Figure S71).No impact on the diastereoselectivity was observed from either study (ESI section 2.11).
The thought occurred that, whilst the BF 4 À anions were located around the Pd(II) nodes in the solid-state, in solution the concentration of strongly hydrogen-bond accepting DMSO molecules would be orders of magnitude higher.The idea that solvent molecules, rather than anions, interacting with the coordination sphere of the cages were responsible for stabilisation of the cis assemblies was thus examined experimentally.
The dilution studies were revisited using CD 3 CN-a weaker hydrogen bond acceptor-as titrant (DMSO and MeCN have hydrogen bond acceptor parameters, β, of 8.9 and 4.7, respectively [31] ).In this instance, the proportion of the trans isomer of C1 P increased slightly with dilution (Figure S81 and S82).Meanwhile, for C1 Q , the percentage of the cis cage decreased; as the proportion of CD 3 CN increased, the minor trans isomer became more prominent (Figure S77 and S78).At a solvent ratio of 9 : 1 CD 3 CN/d 6 -DMSO, the percentage composition of trans-C1 Q actually superceded that of the cis isomer (Figure 3d).Similarly, addition of D 2 O (β = 4.5) [31] to a d 6 -DMSO solution of C1 Q also resulted in enhancement of the trans isomer at the expense of the cis (Figure S72 and S74).
It was concluded that, in the absence of additional effects, the trans isomer is favoured for both C1 P and C1 Q purely on grounds of steric hindrance.The cis isomers, however, provide a suitable site around the coordination sphere for interacting with hydrogen bond acceptors (Figure 3a).Consequently, employing as solvent DMSO-a strong hydrogen bond acceptor-led to a reduction in the relative energy of the cis isomer.For C1 Q this effect was more pronounced due to the four polarised aromatic CÀ H bonds (H b and H j ) around each Pd(II) ion (compared to just two for C1 P À H b ).
The difference in isomer selectivity between C1�NO 3 À and C1�Cl À is proposed to arise partly from a reduction in Pd•••Pd distance, with the smaller guest inducing an increased helical pitch. [32]This increased the offset of bulky donor groups in the trans isomer, further reducing its relative energy, observed as an increase in diastereoselectivity towards trans-C1.For C1 Q , the conformational changes upon guest exchange may also impact the complementarity between the external binding pocket and solvent molecules, reducing effective stabilisation of the cis isomer.This was suggested by a lack of shift in the quinoline resonance H j for trans-C1 Q (Δδ = 0.01 ppm), in contrast to cis-C1 Q (Δδ = 0.11 ppm), upon increasing CD 3 CN composition (Figure 3b-d), indicative of H j being incapable of significant interactions with solvent molecules in the trans cage isomer.
From the combined data, we have constructed a molecular-level picture of the multiple interactions that influence the observed diastereoselectivities between cis and trans cages.The ligand design, endohedral interactions between host and and exohedral interactions with solvent molecules all contribute to the relative energies of the diastereomers.Thus, both first and second coordination spheres [33] play an important role in directing the self-assembly process.Modulation of these factors enables control over isomer selectivity, and opens up the possibility for stimuli-responsive switching of the equilibrium position within isomer libraries. [34]More detailed investigations into this effect are underway and will be reported in due course.

Larger Pd 2 L 4 cages
To probe the utility of these designs with alternative ligand scaffolds, ligand L2 Q (Figure 4a) was investigated.We have previously reported the self-assembly of ligand L2 P ; in d 6 -DMSO a mixture of the cis-and trans-Pd 2 L 4 cage (C2) isomers formed, whilst in CD 3 CN trans-C2 P formed essentially exclusively. [15]It had been suggested that this behaviour arose from the higher polarity DMSO solvent stabilising the more polar cis isomer, [35] without being able to provide a more detailed explanation.In light of the new investigations with C1, it is now proposed that the stronger hydrogen bond acceptor nature of DMSO, compared to CH 3 CN, leads to enhanced stabilisation of the cis-C2 P isomer specifically through hydrogen bonding interactions between solvent molecules and the exohedral face of the Pd(II) coordination sphere.These interactions are less favoured with the trans isomer which, excluding other factors, provides the least sterically hindered primary coordination sphere.
Self-assembly of L2 Q with Pd(II) (as the BF 4 À salt) in d 6 -DMSO resulted in what, superficially, appeared to be nearquantitative (estimated at 70 % by NMR integration; Figure S114) formation of a single species, C2 Q (Figure 4b).In combination, the high-symmetry NMR spectra (Figure 4b and S104-106), ESI-MS (Figure S115-118) and NOESY (H S112) data identified C2 Q as either cis-or trans-Pd 2 (L2 Q ) 4 .The absence of prochiral units within the ligand structure prevented the use of diastereotopic splitting (or lack thereof) as a diagnostic tool to differentiate the two isomers in solution.
The solid-state structure of C2 Q was determined by SCXRD and revealed, as expected, a cis arrangement of ligands within the assembly (Figure 4c-e). [29]Consequently, it seemed the preference under these conditions for cis and trans coordination environments when pairing pyridine with quinoline and picoline, respectively, holds for different ligand scaffolds.

Pd 3 L 6 'double-walled' triangles
To explore higher nuclearity systems, L3 Q and L3 P (Figure 5a) were synthesised.Symmetric, dipyridyl analogues of these ditopic ligands [26] have been shown to assemble into 'double-walled' Pd 3 L 6 triangles (Figure 5b). [36,37]For such structures assembled from an unsymmetrical ligand, one instance of which has been reported by Chand and coworkers, [38] there are 9 possible isomers.Two of these provide all-cis or all-trans arrangements of donors at the three metal nodes (Figure 5b).From our understanding of isomer selectivities induced by CSE, it was hypothesised

Research Articles
that self-assembly of L3 Q with Pd(II) in DMSO would favour formation of the cis-Pd 3 L 6 isomer of C3, whilst L3 P would be biased towards the trans assembly.
C3 Q exhibited a set of dominant signals in the 1 H NMR spectrum (Figure 5c).The diffusion coefficient of this major species, derived from DOSY, corresponded to a solvodynamic radius of 15 Å, whilst isotopic patterns observed by ESI-MS were consistent with assemblies possessing the anticipated Pd 3 L 6 formulation (Figure S139 -146).This major species was estimated by integration of 1 H NMR signals to constitute � 35 % of the isomeric mixture (Figure S138), over three-fold that of a statistical library (11 %).As with the previous systems, the high symmetry of the NMR spectra and through-space interactions observed by NOESY were consistent with only the all-cis or all-trans isomers (Figure 5b).
Weakly diffracting crystals were grown that required the use of synchrotron radiation to obtain satisfactory SCXRD data. [29]The solid-state structure (Figure 5d-f), however, unambiguously revealed each pair of ligands along the three "walls" to be aligned parallel, with a 'head-to-tail' arrangement of the ligands on each face of the triangle.This gave, at each of the three metal nodes, the anticipated cis arrangement of quinoline and pyridine donors.
In contrast to L3 Q , the equilibrated self-assembly mixture of L3 P and Pd(II) in d 6 -DMSO resulted in a 1 H NMR spectrum that appeared to show formation of a mixture of assemblies, with no clear dominant species.Given the established impact of solvent on the equilibrium between cis and trans isomers, self-assembly with L3 P was re-examined in CD 3 CN.This yielded a much simpler 1 H NMR spectrum (Figure 5g), with a clear dominant species ( � 25 % of mixture, Figure S175).
The major product was of high symmetry, as observed by both 1 H (Figure 5g) and 13 C NMR (Figure 5h) which, combined with NOE interactions between the pyridyl and picolyl donor groups (H b •••H l ; Figure S174) and ESI-MS data (Figure S176-186) again indicated formation of either the cis-or trans-Pd 3 L 6 assembly as the major species.Despite multiple attempts, single crystals suitable for X-ray diffraction could not be obtained in our hands and the lack of prochiral units prevented the use of NMR spectroscopy to distinguish between the two possible isomers.Consequently, we turned to DFT to investigate the relative energies of the cis and trans assemblies.
Unexpectedly, use of different functionals for the geometry optimisations resulted in a switching of the cis (HSE) or trans (PBE0) C3 P isomer being lower in energy (ESI section S3).It has been demonstrated how environmental perturbations significantly impact the stability of the individual isomers.Thus, without the suitable inclusion of explicit encapsulated and exohedral solvent molecules and anions within these models, the balance of calculated energies between isomers can be easily swayed.Based on the experimental data obtained for C1, and the DFT calculations previously performed on the more rigid C2 systems, it is tentatively suggested that the most likely identity of the major isomer of C3 P is the trans assembly.

Heteroleptic self-assembly
Preliminary investigations to extend these CSE designs to Pd 4 L 8 'double-walled' tetrahedra [37,39,40] assembled from ligands L4 (Figure 6a)-synthesised through copper-free Sonogashira couplings [41] -proved prohibitively difficult, with complex NMR spectra obtained from equilibrated mixtures with Pd(II) (ESI Section S2. 22 and S2.24).This is perhaps unsurprising.In a statistical library, each of the 35 possible isomers [42] would constitute < 3 % of the mixture, with even the most symmetrical cages (i.e.cis and trans) possessing two ligand environments.Without near quantitative selectivity, identification of NMR signals for a particular isomer would likely be an insurmountable challenge.Consequently, further exploration of these systems was not attempted.
Inspired by a recent report of a heteroleptic cis-Pd 2 L 2 L' 2 MOC, [40] there was motivation to investigate the potential integrative self-assembly between L4 and symmetric ligand L2 H . From such a mixed-ligand assembly there would be two possible isomers: the syn and the (chiral) anti isomer, with the pair of L4 ligands arranged in the same or opposite directions, respectively (Figure 6c).
After equilibrating a stoichiometric mixture of L2 H and L4 Q with Pd(II) in CD 3 CN, DOSY (Figure S224) and ESI-MS (Figure S226-230) confirmed formation of heteroleptic assemblies with the anticipated Pd 2 (L2 H ) 2 (L4 Q ) 2 formulation.Meanwhile, NMR analysis demonstrated the presence of two spectroscopically similar species in an approximately 3 : 1 ratio (Figure 6b and S225).
The lack of bilateral symmetry in L4 Q induced a lowering of the symmetry of L2 H in both cage isomers, resulting in distinct signals for all 12 protons, corroborated by the 13 C NMR spectrum (Figure S211).Whilst analysis of the 1 H NMR spectrum was made challenging by significant signal overlap, 2D NMR techniques (COSY, HMBC and TOCSY) enabled assignment of all peaks for both isomers (Figure S205).
NOESY and ROESY were employed to determine the identities of the two isomers (Figure 6c).Observation of particular through-space interactions ( ) for the minor species (Figure 6d) led to the conclusion that this was the less sterically congested anti isomer.The more limited NOE interactions observed for the major assembly demonstrated that this was the syn isomer, with both L4 Q ligands arranged parallel.This result further supported the conclusion that, rather than selectivity towards the cis isomers of C1À C3 being purely driven by repulsive steric effects, additional stabilising interactions promoted this ligand arrangement.In this instance, presumably the two different second coordination sphere sites around the Pd(II) ions in the syn isomer provide more favourable interactions with solvent molecules compared to those of the anti cage.
The anticipated structure of syn-Pd 2 (L2 H ) 2 (L4 Q ) 2 was confirmed by SCXRD, with both L4 Q ligands arranged parallel to each other (Figure 6e-g). [29]The solid-state structure of the heteroleptic cage was found with one BF 4 À anion encapsulated within the cavity, and external counterions interacting with the external face of the coordination spheres around the two Pd(II) ions (Figure 6e).
The integrative self-assembly between L2 H and L4 P was also attempted.Whilst ESI-MS (Figure S233-241) demonstrated the presence of the heteroleptic assembly (alongside minor signals for the homoleptic Pd 2 (L2 H ) 4 ), signal resolution in the NMR spectrum was insufficient to enable effective analysis (Figure S232).
We have previously been able to arrange two different unsymmetrical ligand scaffolds in defined relative orientations through covalent tethering, forming pseudo-heteroleptic MOCs. [43]Using CSE, we now demonstrate the ability to assemble truly heteroleptic MOCs, derived through integrative self-assembly of ligands with the same denticity, incorporating unsymmetrical scaffolds in an orientationallyselective manner. [44] We have prepared a range of unsymmetrical ditopic ligands, with varying backbone scaffolds, incorporating a pyridine donor paired with a sterically bulky quinoline or picoline moiety, and investigated their self-assembly into Pd n L 2n architectures.This coordination-sphere engineering approach was successful in biasing self-assembly towards specific isomers from a statistical library.Interestingly, quinoline and picoline units promoted cis and trans arrangements of donors at the metal centres, respectively, resulting in diastereoselectivity of the self-assembly process towards different isomers of particular architectures.The ability to use this relatively subtle difference to target the formation of specific metal-organic assembly isomers provides a nuanced approach towards directing the self-assembly of unsymmetrical ligand scaffolds.
After probing the source of this difference in selectivity, it was concluded that interactions between solvent molecules and the exterior of the cage around the metal nodes play a crucial role in determining the relative stabilities of isomers.Thus, not only is the first coordination sphere important in directing the self-assembly, but second coordination sphere effects play a critical role and can, in fact, supercede the directing effects of the primary structure.This insight further opens up the potential for designing stimuli-responsive, [45] shape-shifting systems [46] that respond to changes in their environment.
The orientationally selective incorporation of an unsymmetrical ligand scaffold into a heteroleptic M 2 L 2 L' 2 MOC was also demonstrated.Consistent with the homoleptic assemblies, the major isomer that formed had both bulky quinoline donor units coordinating to the same Pd(II) ion.This result further supported the hypothesis that isomer selectivity can be affected by interactions beyond those simply between components within the covalent structure.
This flexible strategy adds a new approach to preparing metal-organic hosts with increased anisotropy to the metallosupramolecular chemist's toolbox.The continued development of methods to access more structurally sophisticated metal-organic cages [6] will lead to supramolecular hosts exhibiting higher-level behaviours, [7] reminiscent of the impressive properties of natural architectures, like enzymes, that have long provided a source of inspiration for chemists.

Figure 5 .
Figure 5. a) Ligands L3 Q and L3 P ; b) schematic representations of Pd 3 L 6 'double-walled' triangle isomers with all-cis and all-trans donor arrangements; c) 1 H NMR spectrum (500 MHz, d 6 -DMSO) of C3 Q ; SCXRD structure of cis-C3 Q showing d) cis-C3 Q �DMF 2 DMSO, and space filling representation e) from the top, and f) from the side; g) 1 H NMR spectrum (500 MHz, CD 3 CN) and h) 13 C NMR spectrum (126 MHz, CD 3 CN) of C3 P .