Introduction

The compositional tunability of metal-organic frameworks (MOFs) that take advantage of the inherent porosity and large surface areas by interconnecting inorganic metal nodes and multitopic organic ligands has provided a diverse opportunity for tailoring their physicochemical properties. By combining different metal ions and organic ligands, various framework matrices can be formed that display a variety of pore sizes, channel types, shapes, and functionalities1,2,3. The versatility of these materials can be further enhanced by functionalization of the components before assembly or post-synthetic modification (PSM)4 techniques that enable the fabrication of porous materials with high compositional and structural complexities. MOFs have found a number of promising applications such as chemical separation5,6,7, gas storage8,9,10,11, heterogeneous catalysis12,13,14, molecule sensing15,16, or ion conduction17,18. A serious weakness of many MOFs that often limits their utility in practical and industrial applications is that they undergo phase transformations and decompositions. Water is particularly damaging for many MOF materials19,20,21,22. For example, the iconic materials MOF-523 and HKUST-124, consisting of Zn4O(bdc)3 and Cu3(btc)2 (H2bdc = 1,4-benzenedicarboxylic acid; H3btc = 1,3,5-benzenetricarboxylic acid), were extensively examined for the aforementioned applications but were found to be too unstable. The exposure of MOF-5 to humid air under ambient conditions gives rise to a transformation of the high-surface-area MOF-5 to low-surface-area MOF-69c or MOF-5W25,26. Several strategies were tested for enhancing the hydrolytic stability of MOFs:27 (i) the organic ligands were modified to be more hydrophobic28,29,30,31, (ii) tri- or tetra-valent metal cations such as Fe3+, Al3+, Cr3+, and Zr4+ were employed to strengthen the metal–ligand bonds32,33,34,35,36, (iii) nitrogen-containing ligands were engaged with low-valent metal ions37,38, (iv) interpenetration systems were designed to stabilize MOF structures39, and (v) ligands were rigidified to strengthen the stability of the framework40.

Based on a combined computational and experimental exploration, we found that the stability of MOFs can be increased significantly using a simple mechanical approach to strengthening the MOF skeleton. Truss beams have been used in building and furniture construction since ancient Egyptian ages to stabilize rectangular constructs. We hypothesized that a molecular truss beam that can reach across the corner angles of the MOF cage may increase the structural integrity in a similar fashion, as conceptualized in Fig. 1. To the best of our knowledge, this simple and general approach has not been explored for enhancing the stability and durability of MOFs. To test the validity of this approach, we constructed a MOF that shares the same structural skeleton with MOF-5, but one ligand carries a functional group such as a carbazole that can reach across the corner angle to the other ligand in the cage and function as a molecular truss beam. Specifically, 2,5-bis(n-(9H-carbazol-9-yl)-propoxy)-1,4-benzenedicarboxylic acid (CZn-H2bdc) and 1,4-benzenedicarboxylic acid (H2bdc) ligands were used in combination with Zn2+ ions to build the structurally stabilized MOFs.

Fig. 1: Schematic illustration of installing a molecular truss beam to enhance hydrolytic stability of metal-organic frameworks.
figure 1

a MOF-5. b CZn-MOF-5 by shielding the zinc oxide site and rigidifying the frameworks in a combination of 1,4-benzenedicarboxylic acid (H2bdc) and 2,5-bis(n-(9H-carbazol-9-yl)-propoxy)-1,4-benzenedicarboxylic acid (CZn-H2bdc) ligands (n = 3 and 10).

Full size image

Results and discussion

Design and Modeling of Structurally Stable MOFs

We chose a typical isoreticular cubic nanocrystal MOF-5 that consists of multiple frames with zinc oxide tetrahedral units, (Zn4O), connected by 1,4-benzenedicarboxylic acid (H2bdc) ligands forming an alternating chain where each phenyl group is twisted by 90° and their lattice has a [Zn4O(bdc)3]8 stoichiometry in a unit cell23. MOF-5 is known to have no coordinatively unsaturated metal sites, unlike HKUST-1 that adopts a Cu–Cu paddle-wheel structure and features open coordination sites that can bind small molecules such as H2O, EtOH, MeOH, and MeCN41. Nevertheless, the hydrolytic stability of MOF-5 is much lower than that of HKUST-1 due to the pliable structure of the weak Zn–Obdc bonds that allow for water molecules to attack the Zn sites, followed by phase transformation or structural collapse26,42. Therefore, both blocking the water from accessing the metal sites and strengthening the structural rigidity of MOFs may enhance the hydrolytic stability of MOF materials. Previously, various functionalized H2bdc derivatives with nonpolar aliphatic groups were used to improve the hydrolytic stability by installing hydrophobic surfaces within the MOFs28,43,44,45. In particular, the incorporation of short functional groups is attractive because the branches have little impact on the porosity and gas adsorption capability of the MOF materials. Aliphatic groups were found to be less effective43,44.

To enhance both the structural stability and hydrophobicity of MOFs, we took inspiration from building and furniture construction where rectangular angles are structurally stabilized by truss beams, as shown in Fig. 1. After some iterative computational and experimental explorations, we concluded that a promising strategy for installing a molecular truss beam is to combine two different ligands, where only one carries functional groups that are designed to reach across the rectangular angle of the MOF cage to strongly interact with the other ligand.

Exploratory studies revealed that carbazole functionalities that can be easily installed on the MOF-skeleton are well-suited for our strategy, and Fig. 2 illustrates the shapes of the frontier orbitals of H2bdc and CZ3-H2bdc ligands. The lowest unoccupied molecular orbital (LUMO) of H2bdc and CZ3-H2bdc is simply a π*-orbital, and the highest occupied molecular orbital (HOMO) of the CZ3-H2bdc ligand is a π-orbital derived from the pendent carbazole group. Thus, the pendent carbazole group of the CZ3-H2bdc ligand can act as a π-donor, whereas the central phenyl groups of the H2bdc and the CZ3-H2bdc are π-acceptors. Therefore, the CZ3-H2bdc ligands can act as a molecular truss beam by reaching across the corner and forming a stable face-to-face π–π interaction.

Fig. 2: Isosurface plots (isodensity value = 0.05 a.u.) of HOMO and LUMO of the H2bdc and CZ3-H2bdc ligands.
figure 2

Electronic energies (eV) are in parenthesis. Non-essential hydrogens are not shown.

Full size image

Quantum chemical calculations indicated that the CZn-H2bdc ligand is particularly effective when used in combination with the unfunctionalized H2bdc ligand. As highlighted in Fig. 3, the arene functionalities of the carbazole moiety in CZn-H2bdc can engage the arene group of the neighboring H2bdc ligand to form strong π–π interactions, which effectively serves as the truss beam that we imagined. The length of the alkyl tether with which the carbazole is attached to the H2bdc ligand can be varied easily, and our calculations showed that a three methylene tether is ideal (i.e. CZ3-H2bdc). In addition to providing structural support that structurally stabilizes the MOF skeleton, these truss beam constructs afford hydrophobic conditions and block the channel that could be used by water molecules to approach the metal site.

Fig. 3: Computed structures in a unit cell.
figure 3

a MOF-5. b CZ3-MOF-5. c Non-covalent interaction (NCI) plots of the MOF-5 and the CZ3-MOF-5. All bond lengths are in Å. The surface of the NCI interactions is colored on a green scale. Non-essential hydrogens are not shown.

Full size image

To test the proposed construct, computer models were constructed employing periodic boundary conditions using CZ3-H2bdc ligands based on the MOF-5 unit cell structure ((Zn4O)8(bdc)24) by considering intermolecular interactions such as steric hindrance, van der Waals forces, hydrogen bonding, and noncovalent interactions according to Eq. 1:

$$\left( {{{{\mathrm{Zn}}}}_4{{{\mathrm{O}}}}} \right)_8\left( {{{{\mathrm{bdc}}}}} \right)_{24} \,+\, x{{{\mathrm{CZ}}}}3\hbox{-}{{{\mathrm{bdc}}}}^{2-} \to ({{{\mathrm{Zn}}}}_4{{{\mathrm{O}}}})_8\left( {{{{\mathrm{bdc}}}}} \right)_{24-x}\left( {{{{\mathrm{CZ}}}}3\hbox{-}{{{\mathrm{bdc}}}}} \right)_x \,+\, x{{{\mathrm{bdc}}}}^{2-}$$
(1)

As shown in Fig. 3, the structures of [(Zn4O)8(bdc)24], MOF-5, and the designed [(Zn4O)8(bdc)16(CZ3-bdc)8] were geometrically optimized with density functional theory based tight-binding (DFTB) methods that are suitable for treating very large systems. Interestingly, we found that two major interactions support the truss beam forming interactions: (i) a strong π–π interaction between the arene functionalities of the carbazole and the H2bdc moieties, (ii) a non-classical hydrogen bonding interaction between the carboxylate oxygen of the ligand and the alkyl-hydrogen of the tethering group of the carbazole in the CZn-H2bdc ligand. The non-classical hydrogen bond has sufficient influence over the molecular structures46,47, and its bond strength is estimated to be in the range of 2–3 kcal mol–1. The hydrogen bonding (C–H⋯Obdc) with a bond length of ~2 Å gives assistance to the orientation of carbazole groups toward the tetrahedral zinc oxide cluster that is an obvious water-vulnerable site. More significantly, a substantial anchoring effect is exerted by a π–π interaction at a phenyl-phenyl distance of 3.685 Å, tightly tethering the carbazole group that can reach across to the other ligand (See Fig. 3b). These noncovalent interactions (NCIs) were further confirmed (Fig. 3c and Supplementary Fig. 1)48. Thus, these calculations suggested that structural stability is significantly enhanced, which should allow for conserving the compositional integrity of the MOF under humid conditions.

Synthesis and Characterization

Under solvothermal reaction conditions, self-assembly reactions between Zn(NO3)2⋅6H2O and H2bdc in N,N-diethylformamide (DEF) afforded transparent microcrystals of MOF-5. The CZ3-MOF-5 materials were prepared following the identical protocol by combining the Zn sources with a mixture of CZ3-H2bdc and H2bdc ligands (See Fig. 4). Although the crystals differ notably in color, they showed the same cubic crystal morphology owing to the identical connectivity and similarity of the dicarboxylate-based building blocks, as shown in Fig. 4a, b.

Fig. 4: Experimental characterizations of samples.
figure 4

Optical microscope images of a MOF-5 and b CZ3-MOF-5. Diffuse reflectance UV–vis absorption spectra of c H2bdc, MOF-5, and d CZ3-H2bdc, CZ3-MOF-5. e FT-IR spectra of MOF-5, CZ3-H2bdc, and CZ3-MOF-5. f 1H NMR spectrum of CZ3-MOF-5 dissolved in a mixture of DCl/D2O and DMSO-d6 solution. Inset: A peak integration ratio of the 1H NMR signals for the ligands. Scale bars, 0.1 mm (a) and 0.1 mm (b).

Full size image

To confirm the computationally predicted installment of the molecular truss beam, we sought to identify the π–π interaction in CZn-MOF-5 using diffuse reflectance ultraviolet-visible (UV-vis) spectroscopy (Fig. 4 and Supplementary Fig. 2). The UV–vis absorption of the MOF-5 is dominated by the signature of the phenyl group of the H2bdc ligands. On the other hand, the CZ3-H2bdc ligands showed a dyad spectrum based on a combination of the central phenyl group and the pendent carbazole group49. Notably, the absorption peaks were observed at 290 nm and 326 nm, as shown in Fig. 4d, which are assigned to the aforementioned π-donor-acceptor interaction in the CZ3-MOF-5. To further confirm this key interaction, solid-state Cross-Polarization Magic Angle Spinning Carbon-13 Nuclear Magnetic Resonance (CP/MAS 13C NMR) analysis was employed. The solid-state NMR spectra at a spinning rate of 12 kHz showed the de-shielding and shielding effect of carbons affected by π–π interaction (See Supplementary Fig. 3). Also, we investigated the FT-IR spectra paying special attention to the symmetric and asymmetric carboxylate stretching modes of CZ3-H2bdc ligand at ṽ = 1610 cm–1 and 1600 cm–1, respectively, in the two different MOFs while the free ligand shows C=O stretching at ṽ = 1736 cm–1 (Fig. 4e). The C–N stretching mode at ṽ = 1320 cm–1 due to the carbazole group is clearly visible in the CZ3-MOF-5 spectrum. Finally, 1H NMR studies were conducted (See Fig. 4f and Supplementary Fig. 4). The CZ3-MOF-5 sample dissolved in a mixture of DCl/D2O and DMSO-d6 solution displayed four identical protons in H2bdc at 8.02–8.04 ppm and two identical protons in CZ3-H2bdc are found at 7.24–7.28 ppm, confirming that the CZ3-MOF-5 material was successfully prepared. Comparing the peak integration of the 1H NMR signals of the H2bdc to that of CZ3-H2bdc, a ratio of ~2.3:1 is found, which is consistent with the compositional ratio suggested by our calculations for the designed [(Zn4O)8(bdc)16(CZ3-bdc)8] system. Moreover, for the CZ10-MOF-5 sample, the ratio of H2bdc to CZ10-H2bdc is analyzed to ~3.7:1 due to spatial constraints imposed by the bulkiness of pendent groups of the CZ10-H2bdc ligand (See Supplementary Fig. 4).

Hydrolytic stability test

The MOF-5 and CZ3-MOF-5 samples were exposed to moisture under steam conditions at 90 °C and were examined using powder X-ray diffraction (PXRD) to test their structural integrity, as shown in Fig. 5, Supplementary Fig. 5, and Supplementary Fig. 6. The main peaks near 6.8° and 9.7° of the CZ3-MOF-5 are nearly identical to that of MOF-5, and the overall diffraction patterns are similar to the CZ3-MOF-5 sample giving more intense signals, as illustrated in Fig. 5c and Fig. 5d, suggesting that the isoreticular structural frameworks are similar. These diffraction patterns are in good agreement with our computationally optimized structures that showed practically identical framework configurations, as illustrated in Fig. 5a, b, Supplementary Fig. 7, and Supplementary Fig. 8. The PXRD patterns show that exposure to moisture leads to a structural collapse of MOF-5 within an hour, and a phase transition to MOF-5W takes place, indicated by a diffraction peak that appears at 2θ = 8.8°26. In contrast, the CZ3-MOF-5 microcrystals maintained the PXRD patterns even after 5 days and showed a very small peak that indicated a minor phase transition after 6 days, displaying remarkable stability (Fig. 5d). These observations confirm that molecular truss beams are effective for structurally stabilizing the MOF-skeleton.

Fig. 5: Hydrolytic stability with respect to the exposure time to steam at 90 °C.
figure 5

Perspective views of a MOF-5 and b CZ3-MOF-5 in the supercell (3 × 3 × 3). PXRD patterns of c MOF-5 and d CZ3-MOF-5. The asterisk and yellow marks highlight the appearance of a peak at about 8.8° due to the phase transformation of MOF-5 to MOF-5W.

Full size image

Molecular dynamics simulations

We carried out DFTB molecular dynamics (MD) simulations to better understand the molecular basis for the observed differences in hydrolytic stabilities of MOF-5 and CZ3-MOF-5 microcrystals, including water molecules at 298 K and 1 atm in a periodic simulation box (Fig. 6 and Supplementary Fig. 9). It is challenging to capture chemical events such as bond breaking and structural collapse with standard MD simulations, because these events take place in time scales of several seconds, if not hours, while computer simulations can only probe a much shorter time scale. Nonetheless, a careful analysis of the MD simulations can provide much insight into how water molecules approach the vulnerable sites. Previous studies reported that water molecules interact with the MOF-5 by directly attacking the Zn4O moiety via a ligand exchange mechanism where water replaces one of the coordination bonds of MOF-5 (Zn–Obdc)22,50. Hydrogen bonding interaction between H2O and Obdc sites was also identified as important for the structural degradation.

Fig. 6: Snapshots for the equilibrium state based on the DFTB molecular dynamics simulation after 100 ps.
figure 6

a MOF-5. b CZ3-MOF-5. The number of H2O molecules was determined with current density ~0.998 g mL–1 in the periodic boundary condition. Non-essential hydrogens are not shown.

Full size image

In our MD simulations, we prepare the initial state by filling the void in the MOF with randomly oriented water molecules that match a liquid density of ~0.998 g mL–1, which is then equilibrated for standard conditions. These calculations reveal that the vulnerable metal-ligand contact sites in MOF-5 attract clusters of hydrogen-bonded water molecules, as shown in Fig. 6a. All hydrogen bonding interactions shorter than 2.3 Å that qualify as strong are highlighted in sky blue dash lines51. During the MD simulations, we observed an attack of water molecules on the tetrahedral Zn4O unit to strongly coordinate to the Zn atom with Zn–OH2 bond length being 2.15 Å. Among the hydrogen-bonded network of water molecules, a representative hydrogen bonding interactions (Obdc⋯H2O) that will serve a starting point for attacking the carboxylate moiety of the organic ligand is highlighted in yellow in Fig. 5a. As a result of the interactions between MOF-5 and the water molecules, the Zn–Obdc bonds are notably elongated to ~2.46 Å, and the frameworks are highly distorted (Fig. 6a). In contrast, the Zn–Obdc bonds on the CZ3-MOF-5 are maintained at a relatively short 1.94 and 2.03 Å after 100 ps. The simulation shows that the alkyl-tethered carbazole not only provides structural integrity by maintaining the anchoring effect on the MOF-skeleton but plays a crucial role in repelling water attacks, and discourages the formation of hydrogen-bonded water clusters at the vulnerable sites. These results highlight that the structural support combined with the classical approach of increasing the hydrophobicity is in play synergistically to afford the hydrostability.

N2 isotherms and specific surface areas

The N2 isotherm characteristics of CZ3-MOF-5 was explored next. As mentioned above, decorating the organic ligands with large or bulky functional groups can affect the porosity and the internal surface area and other physicochemical properties. Before experimentally testing a gas adsorption/desorption isotherms, we first examined the electrostatic potential (ESP) map of H2bdc, CZ3-H2bdc, MOF-5 and CZ3-MOF-5 cluster models from each unit cell structure (Table 1 and Supplementary Fig. 10). Previously, more polarizable adsorbents with electrostatic potentials that show higher amplitudes exhibited increased abilities to adsorb gases52. As shown in Table 1, the MOF-5 and CZ3-MOF-5 clusters show similar electrostatic potentials as isostructural frameworks except for the pendent carbazole moiety of the CZ3-H2bdc ligand, which displays a notably higher degree of charge polarization. These calculations indicate that CZ3-MOF-5 features a much more polarizable framework surface than MOF-5, which may induce a more robust interaction with adsorbent molecules.

Table 1 Electrostatic potential (ESP) maps and specific surface areas of MOF-5 and CZ3-MOF-5.
Full size table

Previously, a systematic investigation of the N2 adsorption sites in MOF-5 was performed using both experimental and computational methods53,54. The N2 molecules primarily adsorb on the Zn4O tetrahedral sites in the MOF-5 at low loadings. As suggested by the aforementioned studies, the Zn4O sites of the CZ3-MOF-5 are moderately obscured by the pendent carbazole groups, which reduces the favorable binding sites in CZ3-MOF-5 compared to MOF-5. Consequently, the accessible surface area is expected to be smaller in the trust beam stabilized MOF. Our studies show that the BET and Langmuir surface areas of the CZ3-MOF-5 are 1344 m2 g–1 and 1445 m2 g–1 (Table 1, Supplementary Fig. 11, and Supplementary Table 1), respectively, which is indeed smaller than 3171 m2 g–1 and 3418 m2 g–1, seen for MOF-5, respectively. Thus, the carbazole functionality provided robust MOF structures with great hydrolytic stability and moderate surface areas as a truss beam component. To more improve the surface area, a structurally simpler construct that does not interfere with the binding site of the guest molecules may be preferable.

We developed a rational design strategy for increasing the stability of MOF matrices by installing a molecular truss beam that reaches across a rectangular corner of the MOF-skeleton, employing the same principles that structurally stabilize buildings and furniture. A prototype MOF designed to serve as a proof of principle of this strategy, CZn-MOF-5, showed remarkably high structural integrity evidenced by enhanced water-resistance under steam conditions at 90 °C for 5 days. A detailed computational and experimental analysis suggests that the increased stability is the result of a synergistic interplay between increased strength of the lattice and increased hydrophobicity in the MOF cavity. Whereas the carbazole functionality was convenient for this demonstration, as it can readily engage in a strong π–π stacking interaction across the corner, it obscures the nitrogen binding site and notably reduces the active binding surface compared to the parent MOF construct. Our simple strategy inspired by how truss beams are used in everyday life should be generally applicable for stabilizing all MOF-type structures and offers an orthogonal, potentially synergistic approach to conventional strategies. Of particular interest are molecular truss beams that are structurally simpler and are anchored more tightly, for example using covalent bonds across the corner. Efforts towards preparing such MOFs are underway in our laboratories.

Method

Computational details

All calculations were performed with the density functional-based tight-binding (DFTB) theory55,56 as implemented in the Amsterdam Modeling Suite (AMS) products of SCM. Geometry optimizations and molecular dynamics (MD) simulations were performed using the GFN-xTB model57. The simulated system was a cubic cell with three-dimensional periodic boundary conditions. In the MD simulations, the time step was chosen to be 1 fs, which is relatively short but adequate for the purposes of this study. The simulation time for equilibrium states was 100 ps in 100,000 steps. The number of H2O molecules was determined for a liquid density of ~0.998 g mL–1 in the periodic boundary condition. Non-covalent interaction (NCI) was calculated based on the level of promolecular approximation implemented in NCIPLOT 3.048,58 and the surface of NCIs was visualized by Visual Molecular Dynamics (VMD)59 software. The electrostatic potential (ESP) calculations for H2bdc, CZ3-H2bdc, MOF-5, and CZ3-MOF-5 cluster models were performed by single-point calculations with the PBE-D3/6-31G** level of theory60,61,62 with zinc represented by the Los Alamos LACVP basis set63,64,65 in Jaguar 9.166. The ESP maps were obtained using the Jmol67 software. The HOMO and LUMO energies of the H2bdc and the CZ3-H2bdc were reevaluated by single-point calculations using Dunning’s correlation consistent triple-ζ basis set cc-pVTZ(-f) that includes a double set of polarization functions68.