Highly branched and loop-rich gels via formation of metal–organic cages linked by polymers

1Coordination chemistry typically features bonds that are intermediate in bond energy between covalent bond energy and other non-covalent interaction energies (for example, van der Waals interactions and hydrogen bonding). Such bonds have been used extensively for the formation of supramolecular polymer networks/gels–, metal–organic cages (MOCs)– and metal–organic frameworks (MOFs)–; these important classes of materials feature an array of exciting, complementary properties. Materials that incorporate structural features that blend these classes of materials not only capitalize on their individual positive qualities, but also, by way of synergy, potentially exhibit unprecedented and valuable properties–.

2A key component of any material structure is the network branch functionality, f, which is the average number of bridges that connect network junctions. In gels prepared from flexible polymers, an increase in f leads to a direct increase in the elastic modulus. Existing supramolecular metallogels (for example, based on the coordination of Fe3+ and catechol derivatives and structural analogues,) have single metal atoms at their junctions (Fig. 1a, left), and these metals can typically only bind to 2–3 ligands. Thus, the ability to tune f in these systems is limited. In sharp contrast, MOCs and the junctions of MOFs typically comprise metal– ligand clusters with MiLj junction stoichiometry in which j ≥ i > 1. This augmented stoichiometry and increased junction functionality translates into unique cavity structures, but has little impact on viscoelasticity because MOCs and MOFs are generally rigid materials.

3With these considerations in mind and inspired by MOC synthesis–, we wondered if it would be possible to use a multi-metal–ligand supramolecular assembly to drive gelation and yield gels that consist of MOCs linked together by polymers— referred to as polyMOCs. These gels would feature tunable nanoscale junction structures and an enhanced f (Fig. 1a, right). Such an approach would be distinct from traditional supramolecular polymerizations, that generate point-like junctions (Fig. 1a, left), or the pre-assembly of stable MiLj cages followed by aggregation or weak supramolecular crosslinking of these spectator cages–. To our knowledge, the concept of gelation driven by multicomponent MiLj assembly has been considered in only two reports, both of which focus on materials with a low f. First, we described the synthesis of hydrogels with targeted M4L4 square junctions via the assembly of Fe2+ or Ni2+ ions with bispyridyl tetrazine ligands bound to the ends of polyethylene glycol (PEG) chains. Although gelation in this system was only possible through a multi-metal–ligand assembly, we could not characterize conclusively the putative M4L4 clusters, and we proposed that a mixture of clusters of different size was probably present. Nitschke and co-workers later reported hydrogels with targeted M4L4 pyramidal junctions prepared from the assembly of Fe2+ ions and 4,4′-diaminobiphenyl-2,2′-disulfonic acid ligands bound to the ends of PEG. Small-molecule analogues of these ligands did form the target cages in solution, but the cages were not characterized directly in the analogous gels; uptake and release of small molecules from the materials suggested the presence of cavities with distinct environments. Although these examples are encouraging, tuning and enhancing f to enable unprecedented mechanical behaviours in the polyMOC context has not yet been demonstrated.

4We began this study with two hypotheses. First, we reasoned that the thermal annealing of a mixture of Pd2+ ions and PEG terminated with para-bispyridyl ligands designed to form M12L24 cages, (polymer ligand PL1 (Fig. 1b)) or meta-bispyridyl ligands designed to form M2L4 paddlewheels, (PL2 (Fig. 1c)) would generate polyMOC gels, gel-1 and gel-2, respectively, with junction structures similar to those of the target assemblies. Second, we proposed that the difference in average junction size, and the corresponding number of polymer chains connected to each cluster, would translate directly into changes in f and defects (for example, primary loops in which both ligand ends of a single polymer chain are attached to the same junction (red chains, Fig. 1a)) that would lead to unique mechanical properties. Here we use 1H magic-angle spinning (MAS) nuclear magnetic resonance (NMR) spectroscopy, small-angle neutron scattering (SANS), molecular dynamics simulations and oscillatory rheometry to test these hypotheses. Our results provide direct evidence for cage assembly in polyMOCs and show that gelation driven by a metal–ligand multi-component assembly programmed by small changes in ligand structure offers a powerful means to tune the network structure and mechanical properties. We demonstrate that the structure of gel-1 (Fig. 1a, far right), which features a high f and also a large number of loop defects, can be leveraged to replace defects selectively with functional free ligands. Thus, materials with modified junctions can be produced with little impact on the shear modulus, which is not possible in polyMOC gel-2 with a low f.

5We first confirmed that bispyridine ligands similar to those on the ends of PL1 and PL2 but not bound to a polymer (‘free ligands’ L1 and L2 (Fig. 2)) form the expected M12L24 and M2L4 assemblies, respectively, in the presence of Pd2+. Information from studies with these free ligands and their resulting MOCs is used below to validate the structure of polyMOCs. Exposure of L1 to Pd(NO3)2·2H2O in dimethylsulfoxide (DMSO)-d6 (0.100 M) at room temperature (r.t.) provided a heterogeneous mixture with highly broadened 1H NMR resonances shifted to a higher frequency compared with free L1 (Fig. 2a). This mixture transformed into a clear light-yellow solution on heating for eight hours at 70 °C. The 1H NMR spectrum of this solution contained one set of ligand-based resonances consistent with a highly symmetric nanoscopic assembly (Fig. 2a and Supplementary Fig. 10). The 1H NMR resonances in the aromatic region were shifted to a higher frequency compared with those of L1, and the corresponding chemical shifts were virtually identical to those reported by Fujita and co-workers for a similar system.

6On mixing L2 with Pd(NO3)2·2H2O in DMSO-d6 (0.100 M) at r.t., many sets of ligand-based resonances were observed in the aromatic region of the 1H NMR spectrum (Fig. 2b). On annealing for two hours at 70 °C, this mixture coalesced into a single highly symmetric assembly (Fig. 2b and Supplementary Fig. 12). Annealing for eight hours at 100 °C afforded an identical spectrum, which implies that the assembly is stable under these conditions. The paddlewheel complex was characterized further by high-resolution electrospray-ionization time-of-flight mass spectrometry (HR-ESI-TOF-MS); a dominant species with a mass/charge ratio (m/z) that corresponds to the triply cationic paddlewheel mononitrate was observed (Supplementary Fig. 14). Finally, although the quality of our crystallographic data was low, X-ray crystallography confirmed the connectivity of the M2L4 paddlewheel complex (Fig. 2c).

7Molecular dynamics simulations of the assembly of a simplified L1 without the benzyl alcohol substituent (see the Supplementary Information for simulation details) revealed the formation of large clusters after 1 μs with an average number of ligands per cluster, ȳ, of 40 ± 20 (Fig. 2d). In agreement with simulation results from Yoneya and co-workers,, this result captures the early stages of the assembly process; the target M12L24 cages are not formed in high yield after 1 μs. Figure 2d (bottom) shows a representative M12L24 assembly obtained from the simulation. In the case of L2, the simulations yielded ȳ = 6.3 ± 0.5 after 1 μs with several of the target M2L4 paddlewheels present (Fig. 2e). Thus, our simulations suggest that the M2L4 paddlewheel forms more readily than the M12L24 cage within 1 μs. Collectively, these experimental data and precedents from Fujita and co-workers support the notion that ligands L1 and L2 form the target M12L24 and M2L4 assemblies, respectively, on thermal annealing.

8Next we turned to the formation of polyMOCs gel-1 and gel-2 from polymeric ligands PL1 and PL2, respectively, and Pd2+. Exposure of PL1 to Pd(NO3)2·2H2O in DMSO-d6 at 23 °C resulted in the immediate formation of an opaque gel (Fig. 3a), which suggests the presence of large clusters. The gel was annealed under conditions similar to those used to induce the self-assembly of free ligands; the annealing process was monitored by variable-temperature 1H MAS NMR (VT 1H MAS NMR) spectroscopy (Fig. 3a and Supplementary Fig. 24). Owing to the very broad 1H resonances in the MAS NMR spectrum of gel-1 (Fig. 3a and Supplementary Fig. 24), we could not resolve the spectral changes on thermal annealing. However, the aromatic resonances observed for the annealed material (Fig. 3a, red spectrum) have the same chemical shifts as those observed in solution 1H NMR spectra of L1 assemblies (Fig. 2a), and also soluble coordination polymers formed from mixing PL1 with Pd2+ at a high dilution followed by annealing (Fig. 3a, black spectrum, and Supplementary Fig. 23). Although the majority of junctions in gel-1 could be the target Pd12L24 cages (judging from the chemical-shift consistency with soluble analogues and the symmetric peak shape), we cannot confirm this conclusively from MAS NMR; cage fragments or larger clusters could yield similar spectra.

9SANS experiments were conducted to provide further support for the proposed structure of annealed gel-1 (Fig. 3b). The SANS model that best fit the overall scattering curve (Fig. 3b, inset schematic) is a summed model of a power law at low scattering angles/longer distances, indicative of a long-range polymer network structure, and of the core-chain model at mid-to-high scattering angles, which describes the local gel nanostructure (that is, the polymer-bound junctions). Originally calculated by Hore et al. to describe a solid inorganic nanoparticle surrounded by the attached polymer chains in a nanocomposite system, the core–chain model fits well with the proposed gel-1 structure of Pd12L24 cage junctions within a PEG network. The model fit to the SANS data provides a cage radius of 1.7 ± 0.2 nm with approximately 20 polymer chains emanating from and surrounding the cage core. These values agree well with the expected ∼1.8 nm cage radius reported by Fujita and co-workers, and with the fact that we would expect 24 chains per cage if every cage formed perfectly. These SANS data provide strong evidence that the structure of gel-1 is similar to that proposed above (Fig. 1a, far right).

10Molecular dynamics simulations 1 μs after exposure of PL1 to Pd2+ (see the Supplementary Information for the simulation details) revealed the presence of large clusters (ȳ = 21 ± 6) connected by highly extended polymer chains (Fig. 3c). This average cluster size (that is, the number of bis-pyridyl polymer end groups per cluster) agrees quite well with the experimental value observed by SANS, although we stress that, as for the assembly with the free ligands discussed above, after 1 μs the simulated gel-1 does not reflect the reality of the thermally annealed network. Instead, we use simulations here to calculate f and the number of looped chains for non-annealed networks; these values will be important for comparison with mechanical property data (vide infra). The simulated cluster-size distribution in gel-1 (Supplementary Fig. 26) was quite broad and contained some very large clusters with over 50 ligands (Fig. 3c). Given the relatively short polymer chains that link these clusters, a majority of the network chains (68%) are primary loops (Fig. 3c, red chains). These chains do not contribute to f, which leads to a calculated f of 6.7. Although this value is clearly well below the maximum possible value of 24, it is nonetheless higher than that possible for any traditional supra-molecular metallogel based on point-like metal junctions. As discussed below, this fact remains true although thermal annealing reduces f and induces even more loop-defect formation.

11The properties of gel-2 were strikingly different compared with those of gel-1. First, gel-2 was translucent rather than opaque, which immediately suggested the presence of smaller junctions and a more homogeneous network (Fig. 3d). In gel-2, the MAS NMR spectra revealed a transformation similar to that observed for free ligand L2: on heating for one hour at 70 °C, the ligand-derived resonances coalesced and sharpened into single resonances that mapped closely onto the solution 1H NMR spectrum of the L2-based paddlewheels (Fig. 3d and Supplementary Figs 12 and 25). These data strongly suggest that the network junctions are converted into the target symmetric paddlewheels.

12SANS data further support the structure of gel-2. As with gel-1, the best fit for the overall scattering curve is a summed model of a power law to describe the network structure and the core-chain model to describe the local nanostructure (Fig. 3e). From the fit, the calculated radius of the paddlewheel core in gel-2 is 0.53 ± 0.05 nm, with approximately four polymer chains emanating from each paddle-wheel core. Again, these values agree quite well with what we would expect for a gel-2 network architecture based on the crystal structure of the paddlewheel complex (Fig. 2c) and that four polymer chains should be connected to each junction.

13As with gel-1, we used molecular dynamics simulations of gel-2 to interrogate the network structure at the early stages of formation. In agreement with the data shown for free ligands, which suggest that M2L4 paddlewheels form more readily within 1 μs than M12L24 cages, simulations of the formation of gel-2 after 1 μs revealed a preponderance of the target M2L4 paddlewheel assembly (Fig. 3f). The average cluster size in this case was ȳ = 5.3 ± 0.7 bis-pyridyl groups; the cluster distribution possessed a peak that corresponded to clusters containing four bis-pyridyl groups (Supplementary Fig. 26). As expected for the smaller junction size in gel-2 compared with that in gel-1, only 25% of the polymer chains in gel-2 were loops. The calculated f for gel-2 was 4.8, which is greater than four because of the presence of some large clusters.

14Next, we used oscillatory rheometry to relate the mechanical properties of gel-1 and gel-2 to their network structures. First, the storage and loss moduli (G′ and G″, respectively) of gel-1 (5.9 wt% in DMSO-d6 (Fig. 4a)) were studied. Prior to thermal annealing, the high-frequency G′ was 12 ± 3 kPa (Fig. 4a). Based on the phantom network theory of rubber elasticity, which relates G′ to f and the mass density of elastically active polymer chains,, we estimate an f of 6.9 ± 1.6 (see the Supplementary Information for details of the calculation), which agrees well with the value obtained from simulations (Fig. 3c). Thermal annealing led to a 57% decrease in the high-frequency G′ value to 5.2 ± 0.3 kPa, which corresponds to an f of 4.1 ± 0.1 (Fig. 4a, Supplementary Fig. 29 and Supplementary Table 1). To rationalize this decrease in f observed on annealing, we propose that annealing drives the fraction of very large clusters (that increase f) in the non-annealed gel-1 towards the target cluster size of M12L24 and thus reduces f. Furthermore, because the target M12L24 cages cannot pack effectively around each other with relatively short PEG linkers (compared with the cage size) attached to every ligand, the vast majority of the polymer chains must either bridge the same two cages (double loops) or form primary loops. Although neither type of loop can be measured directly in these materials at this time,, the simulation data discussed above for pre-annealed materials suggest that the percentage of primary looped chains can, indeed, be very high. As we show below, the presence of such a large number of loop defects provides the opportunity to convert some of these defects into functional species through free-ligand replacement, which offers possibilities for functional network designs that cannot be realized in materials with a low f and fewer elastically inactive network defects.

15After annealing, the yield stress of gel-1 dropped by 87% from 2.1 ± 0.8 kPa to 0.26 ± 0.11 kPa and the yield strain decreased from ∼18% to ∼6.3% (Fig. 4b, Supplementary Fig. 30 and Supplementary Table 2). These results further suggest that gel-1 consists of large clusters connected by highly extended PEG chains, the latter of which cannot bear large stresses. In future studies, increasing the PEG chain length could facilitate enhancements in the yield stress in gel-1 analogues with potential decreases in G′ offset by a decreased likelihood of primary-loop formation.

16As expected, the mechanical properties of gel-2 were quite different from those of gel-1. The measured G′ for gel-2 prior to annealing was significantly lower than that measured for gel-1 (3.0 ± 0.5 kPa) (Fig. 4c, Supplementary Fig. 29 and Supplementary Table 1). On thermal annealing, a 37% decrease in the high-frequency G′ value to 1.9 ± 0.2 kPa was observed, which, based on the phantom network theory, corresponds to an f of 2.13 ± 0.02 (Fig. 4c, Supplementary Fig. 29 and Supplementary Table 1). This value is close to the limiting value of f = 2 below which gelation cannot occur. As described for gel-1, we believe that annealing converts the large clusters in gel-2 to the target M2L4. As fewer large clusters are formed initially in gel-2 compared with gel-1 (as observed in the simulations above), the corresponding decrease in G′ on annealing is smaller.

17The strain and swelling behaviours of gel-1 and gel-2 were clear indicators of the emergent bulk properties derived from junction self-assembly. Prior to annealing, the yield stress of gel-2 (2.6 ± 0.4 kPa (Fig. 4d, Supplementary Fig. 30 and Supplementary Table 2)) was similar to that of gel-1 (2.1 ± 0.8 kPa). However, although gel-1 showed an 87% decrease in yield stress after annealing, the yield stress of gel-2 decreased by only 31% to 1.8 ± 0.1 kPa. Furthermore, although the yield strain of gel-1 decreased on annealing (Fig. 4d, Supplementary Fig. 30 and Supplementary Table 2), the yield strain of gel-2 increased from ∼83 to ∼110% (Fig. 4d, Supplementary Fig. 30 and Supplementary Table 2). Gel-2 could withstand a more than 17-fold greater strain than could gel-1. Furthermore, gel-2 absorbed 157 ± 9 times its own weight in DMSO after five days (Supplementary Fig. 31 and Supplementary Table 3). In contrast, the swelling ratio for gel-1 was 23 ± 2. These data suggest that the average mesh size is much larger for gel-2 compared with that for gel-1, and that the junctions within gel-2 are potentially more dynamic. Indeed, when a sample of gel-2 was cut into two pieces, it visibly healed on heating (Supplementary Fig. 32). Cuts in gel-1 did not heal under the same conditions (Supplementary Fig. 32).

18The results above highlight how simple polymeric ligand design and the switch from para to meta bispyridine can translate into vastly different polyMOC properties; gel-1 and gel-2 behave as though they were different classes of materials (roughly covalent versus traditional supramolecular gels, respectively). Given the structure of gel-1, which features a large fraction of loops (68% from simulations of pre-annealed networks and 84% from G′ measurements, an assumed maximal f of 24 and no other defects), compared with that of traditional gels, we wondered if it would be possible to replace selectively these loop defects with free ligands (dangling-end defects) that contain alternative functionality. Usually, the mechanical properties of gels (for example, G′) are extremely sensitive to loop and dangling-end defects that reduce f; the addition of even a small amount of free ligand would dramatically lower G′. Given the large loop fraction of gel-1 and that G′ is less sensitive to f for networks with increased f, we suspected that the incorporation of free ligands into gel-1 could be possible with minimal or no net change in G′ (Fig. 5a). In contrast, for gel-2, for which f is lower (∼2.13) and which has relatively fewer loops (∼46% based on G′ and an assumption of no other defects), the introduction of free ligands should immediately reduce the network connectivity towards the limiting value of f = 2. In this case, G′ should drop precipitously with the introduction of free ligand (Fig. 5b); such behaviour would also be expected in all other traditional gels with a low f and few network defects. This junction-engineering concept of the selective exchange of loop defects with functional dangling ends in a gel, with no net change in G′, would represent a feature of gel-1 that, to our knowledge, has not been demonstrated in a polymer network.

19To explore this possibility, we measured G′ for analogues of gel-1 and gel-2 in which, during the gel preparation (see the Supplementary Information for the procedure), varying fractions of polymers PL1 and PL2 were replaced with 2 equiv. of free ligands L1 and L2, respectively (Fig. 5c, filled squares). As before, these G′ values were used to calculate the f values based on the phantom network theory (Fig. 5c, open squares (see the Supplementary Information for details)). When up to 12.5% of PL1 was replaced with 2 equiv. of L1, G′ and f were virtually unchanged. Although G′ for gel-1 begins to decrease rapidly as more free ligand is added, even with 50% of PL1 replaced (a network concentration of 3.8 wt% or 45 mg ml−1 in DMSO-d6), the material retained a modulus comparable to that of traditional supramolecular metallogels with a substantially higher polymer content (network concentration of 10 wt% or 100 mg ml−1 in water). The f value of gel-1 with 50% free ligand was 2.29, which is similar to that of pristine gel-2 with no free ligand. As predicted, the G′ of gel-2 dropped steeply (by 68 ± 9%) after only 12.5% of PL2 was replaced, which confirms that network gel-2 with a low f is more sensitive to network defects.

20Having established that gel-1 is much less sensitive to free-ligand defects than gel-2, we envisioned that additional functionality could be introduced to gel-1 through defect engineering with a functional free ligand (L3 (Fig. 5d)). Although ligand replacement to introduce functionality has been explored thoroughly in the context of rigid three-dimensional networks (for example, MOFs), the concept of free-ligand addition in place of loop defects in gels is a feature made possible by the structure of gel-1. Indeed, replacement of 12.5% of PL1 with pyrene-based fluorescent ligand L3 during the gel preparation (see the Supplementary Information for details) provided a new polyMOC gel that exhibited blue fluorescence under long-wavelength ultraviolet light. This fluorescence persisted after continuous extraction of the gel with DMSO (∼66-fold excess) for two days (Fig. 5d and Supplementary Fig. 33); no detectable L3 was removed by extraction, which suggested that L3 was incorporated within the junctions of the polyMOC. The G′ of L3-modified gel-1 was within experimental error of the analogous gel-1 with non-fluorescent ligand L1 (Supplementary Fig. 34). This demonstration of ligand replacement in the junctions of gel-1 opens exciting avenues for modular polyMOC synthesis; through the use of different free ligands, a range of mechanically uniform materials with distinct properties could be envisaged.

21Herein we describe a novel class of polyMOC materials that feature self-assembled metal–ligand clusters as junctions connected by flexible polymer chains. A combination of MAS NMR, SANS, simulation and rheometry was used to study the structure and properties of these materials. These studies show that polyMOCs designed from compositionally identical but isomeric precursors can display a wide range of viscoelastic properties that spans from covalent-like gels to dynamic supramolecular gels. We demonstrate that in polyMOCs with large junctions and a high number of loop defects it is possible to replace selectively defects with functional free ligands to imbue the material with a novel function (in this case, fluorescence) without compromising mechanical integrity. Given the vast array of metal–ligand combinations that are known to provide discrete supramolecular assemblies, and the potential to incorporate many of these within the polyMOC paradigm, we anticipate the development of a range of new polyMOCs with robust, dynamic and otherwise unprecedented properties.

22To a 1 dram (3.7 ml) scintillation vial was added 20.25 mg (7.5 μmol) of macromer (1 or 2) and then 210.0 μl of DMSO-d6. In a 2 ml scintillation vial, a stock solution of Pd(NO3)2·2H2O in DMSO-d6 was prepared at a concentration of 22.2 mg Pd(NO3)2·2H2O per 1.00 ml DMSO-d6 (after vortexing for about one minute, a clear orange solution formed). This solution (90 μl) was transferred via micropipette to the solution of the macromer, and gelation was observed immediately, although the gel coloration was inhomogeneous. The headspace of the vial was purged briefly with argon, the vial was sealed and heated at 80 °C for four hours to give rise to a homogeneous light-yellow gel (translucent if derived from PL2, opaque if derived from PL1). The molarity of the macromer in the gel (in this case 24 mM) was determined by dividing the number of moles of the macromer used by the total volume of the gel, which accounts for the non-negligible contribution of the polymer to the total volume. All other methods and materials are described in the Supplementary Information.

23The Supplementary Information also contains complete characterization of free ligands L1–L3, assemblies of L1 and L2 with Pd(NO3)2·H2O and PL1 and PL2 (Supplementary Figs 1 and 22), cryo-transition electron microscopy (cryo-TEM) images of linked cages derived from PL1 and Pd(NO3)2·H2O at high dilutions (Supplementary Fig. 23), VT 1H MAS NMR of polyMOCs (Supplementary Figs 24 and 25), computational details (Supplementary Figs 27 and 28), details of SANS experiments and data fitting, complete rheometry data (Supplementary Figs 29 and 30, and Supplementary Tables 1 and 2), swelling data (Supplementary Fig. 31 and Supplementary Table 3), self-healing studies (Supplementary Fig. 32 and Supplementary Table 3), studies of the effect of the percentage of macromer replaced with free ligands L1 or L3 on fluorescence, G′ and f of the polyMOCs (Supplementary Figs 33 and 34, and Supplementary Table 4).

24The X-ray crystallographic data for the structure of the paddlewheel complex reported in this study are deposited at the Cambridge Crystallographic Data Centre (CCDC) under deposition number CCDC 1423278.