The generation of exquisite and complex architectures from relatively simple molecular precursors through noncovalent interaction is ubiquitous in nature [1]. Over the past decades, the self-assembly of metal-organic cages (MOCs) offers a controllable platform at the molecular level to mimic biological counterparts such as protein receptors and enzymes [2–4], in which the spatial arrangement and directionality of the components are well-defined owing to a thermodynamic control of the error correction of the reversible bonds [5,6]. However, different self-assembling progresses often equilibrate to present a distribution in delicate balance, which is tunable depending on external stimuli [7–9], e.g., guest [10,11], concentration [12,13], solvent [14–16], etc. Substantial effort has been devoted to illuminating the principles governing the self-assembly processes from the same building blocks [17–19] nevertheless, precise prediction in many cases still remains an elusive challenge [5,20,21].

Benefited from electrostatic and hydrogen-bonding interactions, anions have been widely employed in self-organization [22], recognition [23–25] and transformation [26,27]. For instance, anionic templates with shape complementarity and ideal packing coefficient were added to the system to drive the formation of inaccessible complicated supramolecular architectures [15,28]. In addition to the mutual matching host–guest complexation [29], dynamic associations of the counterions around a cationic host with large holes opening to the exterior are common phenomena in solution, but are neglected in most cases. Such anion binding may impose elaborate perturbation on the constitutional dynamic libraries (CDLs) [7,8] in solution, thus resulting in various thermodynamic and/or kinetic binding events to lead to structural regulation [30–32], reminiscent of the induced-fit mechanism observed in biological receptors [33]. Moreover, it is of interest to explore how the supramolecular isomerism in coordination self-assembly can be controlled by anions in different structural orders, considering that it has been extensively investigated in polymeric and cyclic systems [20,34–36], but less studied in cage system, compared other external stimuli such as guests, concentration, solvent, pH, and light [12,17,26,28,37–39].

Here, we report differential assembly and transformation of three MOCs, i.e., one Pd₂L₄ (MOC-34) cage and two (Pd₃L₄)_2-3 supramolecular isomeric cages (MOC-35 and MOC-36), induced by BF₄^‒ or NO₃^‒ anions (Scheme 1). Based on the experiments and theoretical calculations, we find that the host-anion dynamic interactions play a significant role beyond anion template effect in the regulation of thermodynamics and kinetics in cage transformation. The anion-induced interconversion between the smaller Pd₆L₈ and larger Pd₉L₁₂ can tune the confined environment, thus generating distinct host–guest interactions for guest inclusion.

A tridentate torsional compound, tris(4-(pyridin-3-yl)phenyl) amine (L), was synthesized as the organic ligand. Owing to the C₃-symmetric propeller shape and the free twist of N_Py-sites, its assembly with the four-coordination Pd²⁺ is expected to form different cage structures with variable composition (Scheme 1 and Fig. 1). A lantern-shaped cage MOC-34 was obtained by mixing L and 0.5 equiv. Pd(NO₃)₂ or [Pd(CH₃CN)₄](BF₄)₂ in DMSO at room temperature for 2 h. ¹H NMR spectrum presents proton signals corresponding to a Pd₂L₄ cage structure containing the degraded C₂-symmetric ligand, which has been fully characterized by ¹³C, ¹H-¹H COSY, ¹H-¹³C HSQC, ¹H DOSY NMR and ESI-HR-MS (electrospray ionization−high-resolution mass) spectral measurements (Figs. S1–S7 in Supporting information). The single-crystal diffraction analysis (Fig. 1a and Table S1 in Supporting information) confirms that MOC-34 is composed of four bidentate ligands and two Pd²⁺ cations, in which only two N_Py-sites of each ligand participate in coordination with the Pd-centers, leaving one pyridyl arm uncoordinated.

Since the tridentate ligands in MOC-34 show unsaturated coordination, we expect that even larger cage structures could be self-assembled by increasing the feeding ratio of metal/ligand (Pd/L > 1/2). Indeed, a clear solution was obtained by mixing L and 0.75 equiv. [Pd(CH₃CN)₄](BF₄)₂ (Pd/L = 3/4) in DMSO at 353 K for 4 h, and the ¹H NMR spectrum displayed an uniformed proton signal pattern (Fig. S1) experiencing apparent downfield shift when compared with the free ligand, indicating that the three pyridyl arms are C₃-symmetrically coordinated with Pd-centers to constitute a Pd₆L₈ type cage (MOC-35, Scheme 1 and Fig. 1b). The formation of Pd₆L₈ cage structure has been further confirmed by the detailed NMR and ESI-HR-MS spectral measurements (Figs. S8–S13 in Supporting information), in which the single diffusion band (logD = −10.1) in ¹H DOSY spectrum suggests a dynamic radius of 13.8 Å for the cage (Fig. S12), and the neat series of successively charged [(Pd₆L₈)¹²⁺+nBF₄]^(12-n)+ species indicate the presence of the sole MOC-35 product in solution (Fig. S13). The single-crystal structure of MOC-35 (Fig. 1b and Tables S2 in Supporting information) unambiguously revealed an octahedral Pd₆L₈ cage consisted of eight ligands and six Pd²⁺ ions, where the square-planar coordination of Pd-center is accomplished by four Pd-N bonds. The triphenylamine skeleton in L maintains the torsional C₃-symmetry, while pyridine moieties are twisted from planarity to wrap around the Pd-vertices (Fig. 1b). The distance along the diagonal of MOC-35 is measured to be 26 Å, in consistence with above DOSY estimation.

It is a surprise that with the same feeding ratio of Pd/L = 3/4 by using Pd(NO₃)₂ instead of [Pd(CH₃CN)₄](BF₄)₂, a new cage structure was assembled (MOC-36) under similar conditions (Scheme 1), which exhibits distinct and complicated NMR signal profiles as compared with MOC-34 and MOC-35 (Figs. S1 and S14–S16 in Supporting information). The careful analyses of the ¹H-¹H COSY and ¹H-¹³C HSQC spectra suggest that the C₃-symmetry of L ligands in this cage is seriously distorted, and three sets of proton signals are well resolved (Figs. S15 and S16), indicating a much lower symmetry of MOC-36 than the octahedral MOC-35. The ¹H DOSY spectrum of MOC-36 (Fig. S17 in Supporting information) confirms that all proton signals have the same diffusion coefficient (logD = −10.17), giving a dynamic radius of the cage at approximate 16.1 Å, which means MOC-36 has a larger cage size than MOC-35 (32 vs. 26 Å). The ESI-HR-MS spectrum shows a series of sharp peaks (Fig. S18 in Supporting information), exactly corresponding to the charge states of a Pd₉L₁₂ type cage varying from +10 to +5, namely, [(Pd₉L₁₂)¹⁸⁺ + nNO₃]^(18-n)+ species which show perfect matching of the experimental and simulated isotopic patterns.

Unfortunately, we failed to grow its single-crystals suitable for X-ray diffraction although many efforts have been devoted. Therefore, optimization of the cage structure based on energy minimization was carried out, yielding a best fit cage model of tricapped trigonal prismatic geometry (Fig. 1c and Fig. S19 in Supporting information). Six Pd-centers constitute the vertices of a trigonal prism, the rest three Pd-centers serve as the side capping vertices, and twelve tridentate ligands occupy the lateral triangular faces. From the top view, MOC-36 shows a general reuleaux triangle shape of C₃-symmetry; therefore, the size of MOC-36 can be estimated as 32 × 23 Å when viewed along or perpendicular to the C₃-axis (Fig. S19), in good agreement with the result calculated from the ¹H DOSY spectrum.

To our knowledge, the unique cage geometry of MOC-36 has not been seen before in coordination cages. When compared with MOC-35, the two cage geometries are related (Fig. S20 in Supporting information). MOC-36 consists of three quadrangular pyramids formed by three rectangular sides of the trigonal prism with three capping vertices, which are connected to form the reuleaux triangle shape. Similarly, MOC-35 has two quadrangular pyramids glued together face-to-face. The dihedral angles between the pyramids are different in the two cages, with MOC-36 having a larger angle, leaving more space for anion interactions (Fig. S20). This geometric difference may be the reason for the anion-induced cage transformation (vide infra). The low symmetry of the trigonal prismatic MOC-36 makes the three pyridyl arms of each L ligand distinct, distinguishable from the positions to bind to different Pd-vertices (Fig. 1c). This explains why three sets of proton signals are observed in the NMR spectra of MOC-36 (Figs. S15 and S16).

The above experiments showed that the type of cage formed depends on the metal-ligand ratio, with Pd/L ratios of 1/2 and 3/4 leading to MOC-34 and MOC-35/36, respectively. To explore cage interconversions, Pd²⁺ salt, L ligand, and ethylenediamine (en) chelator were added to seize Pd²⁺ from the cage (Scheme 1 and Figs. S21–S35 in Supporting information). Adding Pd(CH₃CN)₄(BF₄)₂ to a solution of MOC-34-BF₄ resulted in a single product transformation to MOC-35-BF₄, while adding Pd(NO₃)₂ to a DMSO solution of MOC-34-NO₃ led to quantitative conversion to MOC-36-NO₃ (Figs. S21–S24). The reverse cage conversion from MOC-35 or MOC-36 to MOC-34 was also observed when L ligand or en was added (Figs. S25–S35 in Supporting information). In such reversible transformation processes, excess Pd²⁺ salt or L ligand had little influence on the directional cage conversions, but excess en chelator destroyed the cage to liberate free L ligand. These results suggest that the Pd-N coordination allows for dissociation and association of the L ligand, but the Pd-L exchange speed is slower than the NMR time scale.

Counter anions are crucial in regulating the assembly and transformation of cage structures to form supramolecular isomers MOC-35 or MOC-36. To study the impact of mixed BF₄^‒ and NO₃^‒ anions on the transformation process, Pd(NO₃)₂ or Pd(CH₃CN)₄(BF₄)₂ was added to the DMSO solution of MOC-34-BF₄ or MOC-34-NO₃, respectively. A dynamic mixture of MOC-35 and MOC-36 was observed using NMR analysis (Figs. S36 and S37 in Supporting information) [7,8]. In contrast, using only one type of anion leads to the formation of one product: BF₄^‒ leads to MOC-35 and NO₃^‒ leads to MOC-36. Both cages remain stable at room temperature for an extended period (Figs. S38 and S39 in Supporting information), but at elevated temperatures, MOC-36 transforms to MOC-35, and not the other way around. MOC-36 remains unchanged below 333 K but slowly transforms to MOC-35 above this temperature (Figs. S40–S48 in Supporting information). It takes 15 days for complete conversion to MOC-35-NO₃ at 353 K (Fig. S44 in Supporting information). NMR analysis shows that the converted MOC-35-NO₃ has a comparable cage size to the self-assembled MOC-35-BF₄ (Fig. S49 in Supporting information), but a stronger host-NO₃ interaction based on the remarkably downfield shifts of H_d in MOC-35 (Fig. S50 in Supporting information), indicating dynamic host-anion interactions underneath the Pd-vertices. Heating accelerates the cage transformation, with a rate constant of 5.02 × 10⁻¹⁰ mol L⁻¹ s⁻¹ at 353 K and an activation barrier of 123.9 kJ/mol. These experiments suggest that MOC-36 is kinetically favored, while MOC-35 is thermodynamically preferred.

The anion effect on the cage isomerization has been further studied by adding [(n-Bu)₄N]BF₄ to MOC-36 (Fig. S52 in Supporting information). Despite BF₄^‒ having weaker binding affinity than NO₃^‒, adding large excess of BF₄^‒ leads to competition with NO₃^‒ to dominate host-anion interactions, resulting in a significantly increased MOC-36-to-MOC-35 conversion rate (Figs. S52–S61 in Supporting information). Adding 180 equiv. BF₄^‒ leads to a complete MOC-36-to-MOC-35 conversion in 72 h at 353 K (Fig. 2a), over five times faster than the conversion without BF₄^‒ anions. At a higher temperature, the conversion is even more rapid, taking only 10 h at 373 K (Fig. 2b and Fig. S58 in Supporting information). The conversion rate at 353 K is calculated to be k = 2.87 × 10⁻⁹ mol L⁻¹ s⁻¹, over five times faster than the rate without BF₄^‒ anions. The conversion activation barrier for MOC-36 is significantly lower at 106.6 kJ/mol under 180 equiv. BF₄^‒ condition (Fig. S61 in Supporting information), indicating a significant role of host-anion dynamic interactions in altering the energy barrier for MOC-36-to-MOC-35 structural transformation with BF₄^‒ anions. However, a reversible MOC-35-to-MOC-36 cage conversion is difficult to achieve, no matter by heating (Fig. S39) or adding excess NO₃^‒ anions (Fig. S51).

More information about host-anion interactions was provided by the single-crystal structural analyses of MOC-35-NO₃ and MOC-35-BF₄ (Figs. 3a and b). Six anions are bound inside pockets underneath the Pd-vertices, forming C-H···O or C-H···F hydrogen bonds with the host through the pyridyl and phenyl rings. Density functional theory (DFT) calculations for fragmental NO₃@[Pd(Py-Ph)₄]²⁺ and BF₄@[Pd(Py-Ph)₄]²⁺ (Py = pyridyl, Ph= phenyl) were carried out based on the single-crystal structures of MOC-35-NO₃ and MOC-35-BF₄, respectively. The total binding energy per anion for MOC-35-NO₃ is calculated to be −545.7 kJ/mol, while for MOC-35-BF₄, it is −583.8 kJ/mol. In MOC-35, the tetrahedral BF₄^‒ shows better shape complementarity than the planar NO₃^‒, although the latter has stronger individual hydrogen bonding. We also calculate the total binding energies per anion for NO₃@[Pd(Py-Ph)₄]²⁺ or BF₄@[Pd(Py-Ph)₄]²⁺ units based on the simulated structure of MOC-36 (Figs. 3c and d), giving significantly different enthalpy values of −595.6 kJ/mol for MOC-36-NO₃ and −551.2 kJ/mol for the postulated MOC-36-BF₄. It is apparent that the NO₃^‒ binding is much more conducive to stabilize the Pd-vertices in MOC-36 than the BF₄^‒ binding, which adequately explains why the cage structure of MOC-36 is easily generated in the presence of NO₃^‒ anions.

The crystal structure analysis and theoretical calculations presented here support experimental observations of anion-induced cage self-assembly and transformation. The tetrahedral BF₄^‒ fit best in MOC-35's vertex pockets, while planar NO₃^‒ fit best in MOC-36's, forming strong host-anion interactions. This controls the stoichiometric cage transformation from MOC-34 to MOC-35 with BF₄^‒ and MOC-36 with NO₃^‒. The Pd-N coordination is intrinsically dynamic, allowing for cage isomerization from kinetically favored MOC-36 with tensive N-Pd-N dihedral angles to thermodynamically preferred MOC-35 with relatively relaxed N-Pd-N dihedral angles. While MOC-36 is stabilized at room temperature by NO₃^‒, heating and adding a large excess of BF₄^‒ can expedite cage isomerization via dynamic guest exchange. Ultimately, MOC-35, with better host-guest fitting to BF₄^‒, will entirely replace the constitutional dynamic library of MOC-35 and MOC-36.

The significant differences in cage size and windows of MOC-35 and MOC-36 could endow them with distinct guest inclusion behaviors [40]. To confirm this, we tested the inclusion of four different aromatic hydrocarbons including phenanthrene-9-carbaldehyde (G1), pyrene-1-carbaldehyde (G2), 4-(diphenylamino)benzaldehyde (G3) and 2,4,5-triphenyl-1H-imidazole (G4) with distinguishable molecular symmetry and aromatic moiety (Fig. 4a).

The study of guest encapsulation was performed in a DMSO-d₆/D₂O = 2/1 (v/v) mixture (Figs. S62–S97 in Supporting information). For MOC-35 (Figs. S62–S74), the proton signals of the host display remarkable shifts, of which the Py protons H_a and H_d generally move downfield while the Ph protons H_e and H_f move upfield (Fig. 4b and Fig. S74). All guest signals are up-field shifted, evidently due to a shielding effect of L center toward the encapsulated guests. The association constants corresponding to the interactions between the four guests and MOC-35 cage were estimated by ¹H NMR titrations (Fig. 4c) [41]. The relative strong host-guest interacitons with K_a values in the 1279~3893 L/mol range are determined. Formation of G⊂MOC-35 clathrates via fast guest exchange has been further verified by ¹H DOSY spectra, which disclose that the diffusion coefficients of the exchanging guests fall in between those of the free guests and the host, showing the weighted average of the coefficients of the free and bound guests [42].

For comparison, MOC-36 has a even larger cavity and open windows, so the guest molecules can undergo much faster guest exchange via the cage windows, thus display weaker binding behavoir with the host. This expection is well supported by the ¹H NMR and ¹H DOSY measurements (Figs. S75–S87). Upon guest titrations, the protons of the host appear to be less affected, while those of guest molecules show slight chemical shift changes (Fig. 4b). The binding constants of aforementioned guests are determined similarly using Bindfit method (Fig. 4c). The relatively smaller K_a values in the 656~1141 L/mol range indicate even fast exchange behavior of the guests and their weaker binding ability towards MOC-36 than MOC-35, which are further confirmed by the ¹H DOSY data with closer diffusion coefficients of the exchanging guests to those of the free guest (Fig. 4c). Such different guest binding behaviors suggest a potential to tune the cage cavities via supramolecular isomerization to meet the demand of separation, catalysis and sensing for varied guests with appreciate kinetics and thermodynamics. It is noteworthy that the cage MOC-34 with small cavity but large windows shows more similar guest encapsulation behavior with MOC-36 than MOC-35, evidented by the relatively small proton signal changes and larger difference in ¹H DOSY spectra (Figs. S89–S97).

In summary, we have described self-assembly of three different types of Pd-based metal-organic cages, MOC-34, MOC-35, and MOC-36. The stoichiometric cage transformations from Pd₂L₄-type MOC-34 to Pd₆L₈-type MOC-35 or unprecedented Pd₉L₁₂-type MOC-36 are well-controlled by anion template induction, while the cage isomerization from MOC-36 to MOC-35 is found to be thermal-driven and significantly affected by dynamic host anion interactions. The experimental and calculation results reveal diversified influence of anion effect to trigger and control kinetics and thermodynamics in cage assembly and conversion processes. The distinguishable guest binding behaviors by virtue of supramolecular cage isomerization have been investigated. This study provides an example that helps to understand the interplay between anion templates and host-anion dynamics in the artificial mimicry of natural anion receptors, which is crucial for the self-assembly and transformation of cage hosts.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by the NKRD Program of China (No. 2021YFA1500401), the National Natural Science Foundation of China (Nos. 21821003, 21890380), and the LIRTP of Guangdong Pearl River Talents Program (No. 2017BT01C161).

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cclet.2023.108477.