The Anisotropic Responses of a Flexible Metal-Organic Framework Constructed from Asymmetric Flexible Linkers and Heptanuclear Zinc Carboxylate Secondary Building Units.

A new porous and flexible metal-organic framework (MOF) has been synthesized from the flexible asymmetric linker N-(4-carboxyphenyl)succinamate (CSA) and heptanuclear zinc oxo-clusters of formula [Zn7O2(carboxylate)10DMF2] involving two coordinated terminal DMF ligands. The structural response of this MOF to the removal or exchange of its guest molecules has been probed using a combination of experimental and computational approaches. The topology of the material, involving double linker connections in the a and b directions and single linker connections along the c axis, is shown to be key in the material's anisotropic response. The a and b directions remain locked during guest removal, whereas the c axis linker undergoes large changes significantly reducing the material's void space. The changes to the c axis linker involve a combination of a hinge motion on the linker's rigid side and conformational rearrangements on its flexible end, which were probed in detail during this process despite the presence of crystallographic disorder along this axis, which prevented accurate characterization by experimental methods alone. Although inactive during guest removal, the flexible ends of the a and b axis linkers are observed to play a prominent role during DMF to DMSO solvent exchange, facilitating the exchange reaction arising in the cluster.

Introduction

Flexible metal–organic frameworks (MOFs)[1−5] are a relatively rare but interesting subset of extended framework materials[6,7] that can undergo structural changes in the presence of external stimuli.[8] Their potential to provide a highly adaptable pore environment, which changes its size and shape to suit the requirements of specific chemistry, makes MOFs particularly attractive as the next generation of materials targeted at difficult separation, catalysis, adsorption, or multifunctional applications (e.g., ferroelectric and nonlinear optical properties).[1,9,10] Recently, several attractive features for the development of future porous materials, in particular, anisotropy and asymmetry, have been identified by Kitagawa.[11] However, in order to develop these new materials, we first need to improve our understanding of how their flexible responses arise.[12] This requires in-depth structural characterization of the MOFs before and after their stimulus-driven transitions, which is often hindered by their inherent porosity, structural disorder, or lack of crystallinity, making a structural solution by traditional direct methods particularly challenging. To overcome these issues, it is possible to combine experimental and computational methods. For example, in the case of MIL-88[13] and its isoreticular compounds,[14,15] which display a large “swelling motion” enabled via a hinge motion, the relatively low quality of the diffraction patterns prevented the characterization of the framework’s dynamics via crystallographic methods alone. Therefore, a combination of simulations (force field constrained optimizations) and powder diffraction refinements were employed to obtain a qualitative structural picture of the dynamical response.[13−15]

In this paper, we present a combined computational and crystallographic study reporting the behavior of the new framework ZnCSA, where CSA (Figure (a)) is the asymmetric flexible linker N-(4-carboxyphenyl)succinamate. CSA was selected, because it combines flexible and rigid characteristics, as it can be considered to have one rigid end, consisting of an aromatic ring, and one flexible end, built from amide and aliphatic functionalities. The inherent flexibility involves sp3 carbons and rotatable bonds, which have previously been shown to produce flexible responses in a range of MOFs.[16−18] Although reported as an organic linker in several extended framework materials, the potential of CSA for introducing flexible responses associated with the distinct ends of the linker has not yet been explored.[19,20] The secondary building unit (SBU) in ZnCSA is a heptanuclear zinc cluster with two coordinated terminal DMF ligands of formula [Zn7O2(carboxylate)10DMF2]. Although MOFs built from the same SBU but incorporating rigid bidentate carboxylate-based organic linkers have previously been reported with the same overall topology,[21−23] the incorporation of the flexible CSA linkers enables ZnCSA to display new dynamical behaviors. The three-dimensional connectivity of ZnCSA could be determined by crystallographic methods, enabling us to describe the connections between the inorganic cluster and organic linker components; however, a subset of the CSA linkers displayed crystallographic disorder of the linker orientation and, therefore, the relative binding of the rigid and flexible ends, making the exact linker environment hard to accurately characterize. In addition, ZnCSA loses single crystal crystallinity upon complete removal of its guest species. These obstacles required additional information beyond crystallography to develop an understanding of the flexible response mechanism of the material. Periodic density functional theory (DFT) calculations were therefore used in conjunction with crystallographic data to guide our understanding of the system. These methods revealed that the topology plays a crucial role in the behavior of the framework, with the framework displaying an anisotropic response depending on the environments of the individual linkers. Postsynthetic single crystal coordinated solvent exchange (SCCSE)[24] of the DMF ligands to DMSO was also performed, resulting in a new material that displayed a significantly different predicted response during its guest removal. Although postsynthetic modification in general has shown great potential in the functionalization of MOFs, most examples involve ligand[25−27] or cation exchange,[28−30] and such SCCSE is relatively rarely reported.[31−33] These exchanges, however, offer a distinct route to control MOF behavior, including the modulation of MOFs’ luminescence properties[34,35] and the tuning of the flexible responses in rigid linker MOFs.[36] Work by Manos et al.[24] has also shown the potential for conducting SCCSE on a flexible material, reporting a wide range of solvent exchanges that result in changes to the overall framework size. Distinctively from these previous SCCSE examples, the mechanism for the exchange reaction at the SBU of ZnCSA appears to be directly linked to the flexibility of the CSA linker. These important structural rearrangements, occurring in the previously locked double linker connections, lead to a significantly different predicted response during guest removal.

Figure 1

(a) CSA linker with distinct rigid and flexible ends. For the single c axis linker, these ends of the molecule are disordered such that it coordinates randomly via either end. Atoms labeled from A to F are involved in the definition of torsion angles φ (Ocarbox(A)–Ccarbox(B)–Csp3(C)–Csp3(D)), ψ (Ccarbox(B)–Csp3(C)–Csp3(D)–Camide(E)), and δ (Csp3(C)–Csp3(D)–Camide(E)–Namide(F)). Note that the values of these torsions are tabulated in Supporting Information S7 for the structures discussed in the text. (b) Secondary building unit of ZnCSA·DMF (1) showing the 7 Zn atoms, 10 carboxylate connections, and 2 terminal DMF ligands. Carboxylate carbons colored green are connections running along the a axis, carbons colored purple are connections running along the b axis, and carbons colored orange are connections running along the c axis. Striped colored connections are connections through the flexible end of the linker, solid colors are connections through the rigid end, and half striped connections correspond to a 50/50 distribution through the flexible and rigid ends of the linkers. The carboxylates showing mixed connections correspond to the single c axis linkers. Note that two b axis connections are hidden behind the cluster. (c) Schematic showing the linker connections between the metal clusters situated at the corners of each crystallographic unit cell. Blue ellipses represent the metal clusters, double green lines represent CSA linker connections along the a axis, double purple lines represent CSA linker connections along the b axis, and single orange lines represent CSA linker connections along the c axis. Disorder has also been removed from the c axis linker for clarity, and coordinated DMF molecules are present on the complete clusters, although their orientations make them hard to distinguish from the Zn polyhedra. (d) Overlay of the three different linker conformations present in 1, colored based on the axis it defines. (e) The two half-occupancy inversion-symmetry-related orientations of the single c axis linker modeled crystallographically. Hydrogen atoms have been removed for clarity.

Experimental Section

General

All reagents were purchased from Sigma-Aldrich Ltd. and were used as received without further purification.

Synthesis

ZnCSA·DMF

A 0.2 M solution of Zn(NO3)2·6H2O in N,N′-dimethylformamide (DMF) (375 μL, 0.075 mmol) was added to a (10 mL) screw capped Pyrex vial containing N-(4-carboxyphenyl)succinamic acid (CSA-H2) (6.9 mg, 0.03 mmol). An additional volume of DMF (3.33 mL) was then added to the mixture, and the vial was then capped and placed in an oven. The temperature inside the oven was raised slowly to 100 °C (1 °C/min), and the reaction was left to proceed for 24 h before cooling back to room temperature at a ramp rate of 0.1 °C/min. The reaction afforded a white crystalline precipitate, which was washed and stored in fresh DMF.

ZnCSA·DMSO

Crystals of ZnCSA·DMF (20 mg) were transferred from DMF into a vial containing dimethyl sulfoxide (DMSO) (1 mL). The crystals were left to sit in the vial for 5 days, with the solvent being exchanged twice daily.

Thermogravimetric Analysis

Thermogravimetric analysis was carried out on a TA 500 instrument in a temperature range of 25 to 600 °C with a scan rate of 3 °C/min and air flow rate of 60 mL/min.

X-ray Crystallography

Single crystal X-ray diffraction data were collected for ZnCSA·DMF using a Rigaku MicroMax-007 HF X-ray generator with a Mo Kα rotating anode microfocus source and a Saturn 724+ CCD detector. Crystals were transferred from DMF into Fomblin oil and then mounted onto the diffractometer using a MiteGen MicroMount. The sample temperature was maintained at 100 K using a 5 mL/min N2 flow from an Oxford Cryosystems Cyrostream Plus device. Intensity data were indexed, integrated, and corrected for absorption using CrysAlisPro software.[37] Single crystal X-ray diffraction data for ZnCSA·DMSO meanwhile were collected using synchrotron radiation at beamline I19, Diamond Light Source.[38] Intensity data, also collected at 100 K, were indexed, integrated, and corrected for absorption using Xia2.[39,40] All structures were initially solved by direct methods in the conventional setting of the unit cell.[41] The atomic positions were analyzed and compared to the other structures collected for the material. Lattice transformations were then applied using the program WinGX to ensure the structures were consistently described throughout the manuscript.[42] Details can be found in Supporting Information S2. The crystal structures in the new unit cell were refined by full-matrix least-squares using SHELXL accessed via the program Olex2.[43,44] As a result of the porous and flexible nature of the material, high resolution data could not always be obtained. Non-hydrogen atoms were therefore only refined anisotropically if there was sufficient data to parameter ratio available. Hydrogen atoms were placed in idealized positions and refined using a riding model with isotropic thermal parameters dependent on the connected atom. In the crystal structures, one of the organic linkers is situated on an inversion center and is therefore highly disordered. Restrained isotropic thermal parameters combined with idealized distance restraints were used to ensure a chemically sensible model for this linker. The routine SQUEEZE from the program PLATON was used to account for the scattering contribution of disordered solvent molecules contained within the large accessible void space of the framework.[45] Full details of the crystals, data collections, and refinement parameters are given in Supporting Information S1. Powder X-ray diffraction data were collected on a Bruker D8 advance diffractometer in transmission geometry using monochromated Cu Kα radiation and 0.7 mm borosilicate capillary tubes. As a result of the sensitivity of the material to loss of solvent, data for ZnCSA·DMF was initially collected while the sample was immersed in DMF within a sealed 0.7 mm borosilicate capillary tube. Crystal structures were visualized and images were produced using a combination of Mercury 4.0 (CCDC)[46] and VESTA 3.[47]

Computational Methods

DFT calculations were performed using the VASP[48] code. Input geometries were generated from experimental structures, in which all atomic positions except for the disordered linker along the c axis had been determined. To account for the experimental random coordination of the c axis linker, 2 × 2 × 1 periodic supercells were built alternating the binding through the rigid and flexible ends of the linkers in the c axis direction. Each supercell contained 20 CSA linkers, 28 zinc cations, 8 cation-bridging oxygens, and 8 DMF or DMSO molecules coordinated to the SBU, a total of 652 or 628 ions for ZnCSA·DMF and ZnCSA·DMSO, respectively. The experimental structures (1, 2, 3) were initially optimized with the unit cell parameters fixed (ISIF = 2) at the measured values, yielding structures C1, C2, and C3 in the respective cases of DMF, partially removed DMF, and DMSO. The flexible behavior of the desolvated materials was then computationally addressed via geometry optimizations by allowing both ion positions and unit cell parameters (volume and shape, ISIF = 3) to relax, thus producing the structures C1 and C3. All calculations were conducted with their coordinated guest molecules but not any guest molecules contained within the pore because of computational cost. Allowing both the unit cell parameters and ion positions to relax therefore probes the structural effect of removal of the solvent from the pores of the material. The ion–electron interaction was described with the projector augmented-wave (PAW) method.[49] To account for van der Waals dispersion forces, the nonlocal correlation functional method was used[50] with the optB86b-vdW exchange functional.[51] An energy cutoff of 520 eV was employed for the plane-wave expansion as well as a k-point mesh of 1 × 1 × 2 to sample the Brillouin zone in the reciprocal space. For all calculations, the “normal” precision setting with convergence criteria of 1 × 10–6 eV for the electronic energy convergence and 1 × 10–5 eV for the ionic convergence were used.

Results

The MOF ZnCSA·DMF was synthesized using solvothermal methods from Zn(NO3)2·6H2O, CSA-H2 (Figure a), and DMF. This afforded plate-shaped single crystals of formula [Zn7O2(CSA)5DMF2]·14DMF, which were analyzed by X-ray diffraction at 100 K. The crystal structure of ZnCSA·DMF (1) revealed that the material crystallizes in the triclinic space group P1 (No. 2) with unit cell dimensions a = 15.7350(7) Å, b = 17.4602(8) Å, c = 19.3022(9) Å, α = 95.955(4)°, β = 109.926(4)°, γ = 105.846(4)°, and V = 4681.2(4) Å3. The framework is constructed from a heptanuclear secondary building unit of formula [Zn7O2(carboxylate)10DMF2]. The cluster contains 7 zinc atoms bridged by 10 bidentate carboxylate groups (8 μ2 connections and 2 μ3 connections), 2 μ4-O atoms, and 2 terminal DMF ligands (see Figure b). The carboxylate groups belong to 10 different CSA linkers, bound through a mixture of rigid and flexible ends, which connect to 6 other clusters to give a non-interpenetrated fqr net.[52,53] The metal clusters are located at the vertices of the crystallographic unit cell with the linker connections running along the crystallographic axes, four corresponding to the crystallographic a axis, four corresponding to the b axis, and two corresponding to the c axis. Within the cluster, there are one octahedral, two square pyramidal, and four tetrahedral Zn environments. The octahedral Zn is located at the center of the cluster and is apically coordinated to the two μ4-O2– anions and equatorially coordinated to four different carboxylate groups. This Zn is then connected through the two μ4-O2– anions to two equivalent trinuclear units, each centered around one of the O2– anions and consisting of one square pyramidal Zn (with the coordinated DMF ligand) and two tetrahedral Zn atoms. The overall cluster could also be considered as two symmetry-related Zn4O tetrahedra, which share the central octahedral Zn as a common vertex. This metal cluster unit has been previously reported in molecular complexes (although often with different terminal ligands),[54−62] one-dimensional coordination polymers,[61,63−66] and three-dimensional metal–organic frameworks.[21−23] In ZnCSA, the carboxylates located around the center of the cluster, coordinated to the central octahedral Zn, are all connections made through the flexible ends of the asymmetric CSA linker and correspond to two a axis linkers and two b axis linkers. The carboxylate groups on outer parts of the cluster meanwhile are connections to the rigid ends of the remaining two a and two b axis linkers, displayed in green and purple in Figure b, respectively, or connections to the two c axis linkers, displayed in orange.

The four a and four b axis CSA linkers can be divided into pairs, each corresponding to double linker connections between two different clusters. Each pair of asymmetric linkers is oriented antiparallel to each other (i.e., the rigid ends of each linker pointing in opposite directions and crystallographically ordered) above and below the crystallographic axis, related by inversion symmetry. The c axis, meanwhile, only shows one linker connection on either side of the cluster, running directly along its axis. This is a key component of the structure and from here will be referred to as the single c axis linker. As with the a and b axes linkers, the single c axis linker binds through both its rigid and flexible ends; however, it is crystallographically disordered around the inversion center because of a random distribution of the two possible orientations within the whole structure. The crystallographic model of these two orientations, which required a number of bond distance restraints, is shown in Figure e. The three linkers modeled in the asymmetric unit, two full occupancy corresponding to the a and b axes likers and one half occupancy corresponding to disordered c axis linker, were all observed to adopt different conformations from each other based on rotations around the torsional angles of the flexible end of the CSA linker. The rigid ends of the three linkers, however, remained the same. This can be seen visually in Figure d, displaying an overlay of the three crystallographically modeled conformations. The torsional angles involved, displayed in Figure a, can be defined as φ (Ocarbox(A)–Ccarbox(B)–Csp3(C)–Csp3(D)), ψ (Ccarbox(B)–Csp3(C)–Csp3(D)–Camide€) and δ (Csp3(C)–Csp3(D)–Camide€–Namide(F)). In particular, the a axis linkers are highly bent, with a ψ value of 60°, whereas the b and c axis linkers are almost planar with ψ values of −176 and 163°, respectively. The values of these torsions in all structures discussed are tabulated in Supporting Information S7. Overall, the framework affords a large (2032 Å3) three-dimensional pore structure with absolute window apertures of 12.7 × 12.6, 12.3 × 9.0, and 4.5 × 4.3 Å. The pore limiting diameter, defined as the smallest opening along the pore, calculated with the Zeo++ software package,[67] is 9.7 Å. The solvent accessible void (measured with a 1.2 Å probe radius) accounts for 43.4% of the unit cell volume. On the basis of the thermogravimetric analysis, there are 14 guest DMF molecules contained within the pore, in addition to those coordinated to the cluster (see Supporting Information S3), which corresponds well to the crystallographic data suggesting 13.6 guest DMF molecules. This value was obtained from a combination of the one guest site in the pore, which could be resolved in the asymmetric unit, giving a refined formula of [Zn7O2(CSA)5DMF2]·1.79(1)DMF, and the residual electron density calculated by the routine SQUEEZE. As a comparison, the absolute pore volume could accommodate 15.8 DMFs, assuming a packing density identical to liquid DMF. A schematic of ZnCSA, designed to emphasize the relationship of the structure to its crystallographic unit cell, is given in Figure c and colored based on the connections along the different axes. More detailed views of the structure along with its Connolly surface can be found in Figure . The bulk phase purity of the material was confirmed by PXRD at room temperature measured while the material was immersed in DMF. The diffraction pattern was indexed to a unit cell of dimensions a = 15.877 (1) Å, b = 17.500(1) Å, c = 19.842(2) Å, α = 97.702(5)°, β = 103.279(4)°, γ = 104.588(4)°, and V = 5083.8(6) Å3, which is a 9% anisotropic expansion compared to the single crystal data collected at 100 K, with the majority difference being in the β angle.

Figure 2

(a) View of the channel running along the c axis in ZnCSA·DMF (left), the double linker connection running along the a axis (center), and the Connolly surface (calculated using a 1.2 Å probe radius) (right). (b) View of the channel running along the a axis (left), the double linker connection running along the b axis (center), and the Connolly surface (calculated using a 1.2 Å probe radius) (right). (c) View of the channel running along b axis (left), the single linker connection running along the c axis (center), and the Connolly surface (calculated using a 1.2 Å probe radius) (right). Angles ω1 and ω2 are shown here but are given in more detail in Figure b. Note that disorder has been removed from the single c axis linker for clarity.

Figure 4

(a) Extracted sections of the overlay of the experimentally determined structure 1, shown using ellipsoids drawn at 50% probability, and the DFT optimized structure C1, represented with a green stick model. Root-mean-square displacements (RMSD) are given below each section. (b) Left: The two half-occupancy inversion-symmetry-related orientations of the single c axis linker modeled crystallographically in 1. Right: Two orientations of the single c axis linker in C1 obtained by artificially introducing inversion symmetry into the non-disordered model post calculation. (c) Environment of the single c axis linker in C1 showing the defined coordination angles θ1 and θ2 and torsional angles φ, ψ, and δ. (d) Overlay of computationally optimized single c axis linkers in C1 (dark colors) and the partially desolvated C2 (light colors). The linkers are overlaid using the two carboxylate oxygens, and the two Zn atoms which they coordinate to, of the flexible end of the linker. Hydrogen atoms have been removed for clarity.

A PXRD pattern of ZnCSA·DMF was also collected once the material was filtered, dried, and heated at 120 °C for 30 min. TGA data (collected before and after heating) suggests that this temperature should be sufficient to remove the majority of the contained guests for the material (10 out of 14 guests) (see Figure a) and, therefore, probes the behavior of the framework during removal of the DMF molecules in the pore but not from the SBU. The large difference observed in the two obtained PXRD patterns (see Figure ) indicates clear changes to the unit cell parameters and, therefore, the MOF crystal structure, during the desolvation process. Unfortunately, as a result of limitations in the data quality and the triclinic nature of the material, this new pattern could not be indexed. However, the general shift to larger 2θ values is suggestive of a reduction in d spacing and therefore a reduction in overall unit cell volume. Heating to higher temperatures to completely remove the guests (beyond point B in Figure a) resulted in a loss of crystallinity. This fully activated material also displayed a negligible volumetric uptake of nitrogen (Supporting Information S6).

Figure 3

(a) Thermogravimetric analysis of ZnCSA·DMF (1) ramping at 3 °C/min. (b) PXRD patterns of ZnCSA·DMF while immersed in DMF (blue) and after drying and heating at 120 °C to remove some of its guests (green). The labels A and B in the figure are to show the position of the obtained PXRD patterns in the solvent loss process. A corresponds to the as-synthesized material 1, whereas B is a partially desolvated material but is significantly more desolvated than the partially desolvated single crystal structure 2.

To gain further structural insight into the framework changes during guest removal, a single crystal structure was obtained after the material was left to dry in air for 1 h, affording the partially desolvated material 2. The level of desolvation of this crystal, however, is unknown, because the significantly reduced data quality from undergoing the phase transition makes analysis of the residual electron density unreliable. 2 showed an identical building unit (metal-oxo cluster and bound DMF molecules), connectivity, and topology to 1 but displayed noticeable differences in the relative angles between the different metal cluster connections, resulting in changes to the unit cell (a = 15.644(1) Å, b = 17.386(1) Å, c = 18.668(3) Å, α = 97.05(1)°, β = 115.49 (1)°, γ = 105.851(7)°, and V = 4236.8(9) Å3; a 9.5% volume reduction). In particular, the single c axis linker changes significantly, which is directly reflected in the change of β from 1 (Δβ = 5.5°). This resulted in a reduced volume for the unit cell and therefore a reduction in void space from 43.4 to 38.3%. The presence of crystallographic disorder on the single c axis linker, requiring a number of bond distance restraints to model, means information on the changes to this part of the material could not be accurately interpreted from the refinement. To approach this problem and provide a chemically sensible model for the disordered components of the two structures, we employed computational methods, optimizing non-disordered versions of crystal structures 1 and 2 using DFT within a fixed 2 × 2 × 1 supercell, C1 and C2, respectively (where C stands for computational, and 1 and 2 stand for the crystal structure used as input geometry). Although computationally expensive, this choice of supercell dimensions (double the experimental cell along a and b axes) allows us to model the crystallographic disorder in an alternating manner, such that for any given SBU, the binding of the single c axis linker (e.g., through its rigid end) is different to its four neighboring SBU’s (e.g., through its flexible end) in the ab plane. It is worth mentioning that computational structures were always modeled with uncoordinated solvent molecules removed (the terminal DMF coordinated to the SBU are preserved); therefore, fixed unit cell (to the experimental parameter values) geometry optimizations (the subscript “opt” refers to optimized) were intended to represent the solvated crystal, but the solvent was not actually modeled using the experimentally determined compositions because of computational cost. The overall validity of the modeling method was assessed by directly overlaying 1 and C1 (using the central atoms of four different metal SBUs) and comparing the positions of the computationally optimized atoms with the non-disordered atomic sites observed experimentally (i.e., the framework without the single c axis linker). As 1 displayed significantly higher resolution experimental data than 2, it was deemed most reliable for this analysis. Figure a shows the extracted sections of the overlay for all the ordered and crystallographically unique components of the structure, including the SBU, one of the two symmetry-related linkers lying along the a axis, and one of the two symmetry-related linkers lying along the b axis. A good agreement between the experimental and computational models was observed, with the positions of all the calculated atoms, represented using a green stick model, lying within the anisotropic thermal displacement ellipsoids determined experimentally. Small root-mean-square displacement (RMSD) values (see Figure a) of the individual section overlays further support the computational–experimental agreement on the non-disordered parts of the structure. Although the disordered components of the structure could not be directly compared in a similar way, the crystallographic model of the disordered single c axis linker and the calculated linker environment after inversion symmetry applied post calculation are shown overlaid in Figure b, which suggests no major differences. Accordingly, the computational models were deemed suitable to provide reliable information and could be used to study the changing single c axis linkers in the structure.

To understand the starting point for this change, the environment of the asymmetric linker in the computed ordered structure of C1 should be described initially. Significantly, this single c axis linker, shown in Figure c, is predicted to lie slightly out of the plane of the connection between its two SBUs. This distortion can be quantified by the two angles ω1 and ω2, one measured on either side of the linker, using the centroid of the two Zn atoms from the connecting SBU, the centroid of the two coordinating carboxylate oxygens, and the center of the linker. These angles are almost identical, with ω1 = 154° at the rigid end and ω2 = 153° at the flexible end of the linker. At the rigid end, ω1 can be seen to primarily arise from the direct coordination angle of the carboxylate group (θ1) at 158°, defined in Figure c using the centroid of the two Zn atoms, the centroid of the two carboxylate oxygens, and the first carbon after the carboxylate group. Meanwhile, at the flexible end, ω2 cannot be described solely by the much wider θ2 at 170°, defined in Figure c consistently with θ1. The same ω (bend out of plane) is instead achieved through additional conformational adjustments driven by rotations of three torsional angles associated with the linker’s sp3 carbons. These angles (φ, ψ, and δ) become −158, 163, and 152°, respectively.

Comparing C1 and C2, to understand the structural transformation upon guest removal, the major change arises due to a shift in the degree that the c axis linker lies out of the plane of the connection of the SBUs, with ω1 and ω2 both changing to become 144°. At the rigid end of the linker, this occurs only through decreasing θ1 to 149°, whereas at the flexible end of the linker, θ2 decreases to 162°, combined with further torsional adjustments to ψ and δ, which change to 158 and 149°, respectively. φ meanwhile is predicted to remain almost the same at −157° (torsion values tabulated in Supporting Information S7). The resulting effect is shown in Figure d, which shows an overlay of the predicted c axis linker in C1 and C2. Although these angles cannot be directly compared to experimental results, as the modeling of disorder makes exact torsional measurements on the individual components unreliable, an average value for the bend of the linker out of plane at both ends can be obtained. This shows a very similar overall response to that predicted by calculation, with the averaged ω changing from 156 to 146°, in very good agreement with the predicted 154 to 144° change. Therefore, although guest species adsorbed within the pores are not explicitly modeled in the calculations, fixing the unit cell to its experimental parameters allows us to reliably reproduce the change in the c axis linker responsible for the material flexible response.

Further treatment of single crystals, including heating at 120 °C to mimic the preparation conditions of the sample from which the powder diffraction pattern was collected, resulted in the complete loss of single crystal crystallinity. Therefore, to try to predict the structural behavior of the MOF during full solvent removal, our DFT calculations were extended by optimizing the empty structure while allowing the dimensions of the 2 × 2 × 1 supercell to vary. This “full structural relaxation”, C1 (“relax” stands for relaxed), in which both the ions and unit cell parameters are allowed to relax, aims to predict the energy minimum of the solvent free framework. This is particularly relevant, because the lattice parameters could not be determined from the PXRD and, therefore, even the size of the repeating unit is completely unknown. The optimization, which gave similar relaxed geometries for both 1 and 2, suggests the material undergoes a drastic change to its unit cell, equivalent to a = 15.30 Å, b = 17.19 Å, c = 15.15 Å, α = 96.8°, β = 138.43°, γ = 109.58°, and V = 2010.4 Å3 (C1) when related back to a 1 × 1 × 1 cell. This new cell is roughly 40% the size of 1 and, similar to the differences between C1 and C2, a result which has been experimentally validated, the change between C1 and C1 is primarily driven by movement of the single c axis linker, producing large differences in the c axis length and β angle. Although this cell does not directly index the powder pattern obtained after heating at 120 °C, the d spacings of the first two predicted peaks (14.1 and 13.1 Å) are similar to those observed experimentally (14.4 and 13.4 Å), therefore suggesting that it may provide a reasonable approximation to the largest changes in the structure. A comparison of the experimental powder pattern after heating at 120 °C and the predicted powder pattern of C1 is shown in Supporting Figure S5. The changes to the structure between C1 and C1 occur broadly through the same mechanism observed in the optimized single crystal structures (C1 and C2) (i.e., changes in the conformation and binding of the c axis linker) but to a larger extent. ω1 reduces to 109°, driven primarily by a reduction in θ1 to 126°, but now also involves a bending out of plane of the aromatic ring. ω2 meanwhile reduces to 114°, θ2 changes to 143°, and the torsional angles φ, ψ, and δ become −115, 164, and 168° (torsion values tabulated in Supporting Information S7). Compared to the change between C1 and C2, C1 shows a much more significant response in the torsion φ during the optimization, accounting for the largest change in the flexible end of the linker. The overall mechanism, shown in Figure , causes a large reduction in the accessible void space to 5.9% and in the pore limiting diameter to 3.7 Å. C1 displays a complete removal of the channels running along the a and b axes, resulting from the single c axis linker folding back to lie almost in plane with the a axis linkers (see Figure d). This anisotropic single axis response observed in ZnCSA during the “structural relaxation” is closely related to the MOF’s topology. The presence of the double linker connections along the a and b axes prevents any significant changes along these directions, thus effectively providing two-dimensional rigid platforms. The single CSA linker connections between these platforms, however, grants the necessary freedom to respond along the c axis, with changes occurring at both its rigid end (change in coordination angle θ1 in a hinge motion) and at its flexible end (conformational adjustments of the torsional angles φ, ψ, and δ).

Figure 5

(a) Computed optimized structure with fixed experimentally determined cell parameters, C1, viewed down the a axis. (b) Computed relaxed structure after full relaxation of both positions and cell parameters, C1, viewed down the a axis. (c) Two SBUs of C1 connected via the single c axis linker connection. φ and θ1 are shown in expanded sections. (d) Two SBUs of C1 connected via the single c axis linker connection. φ and θ1 are shown in expanded sections to illustrate the conformational rearrangements and the coordination change observed at the flexible and rigid ends of the linker, respectively.

The behavior of ZnCSA was further explored through exchange of the DMF solvent in ZnCSA·DMF (1) with DMSO giving ZnCSA·DMSO (3), formula [Zn7O2(CSA)5DMSO2]·18DMSO. The crystal structure was obtained by X-ray diffraction at 100 K and transformed from its standard setting to ensure consistency in the structural description. 3 shows a drastic change in the unit cell dimensions, a = 17.605(4) Å, b = 17.548(4) Å, c = 20.650(6) Å, α = 110.09(2)°, β = 79.19(2)°, γ = 87.58(2)°, and V = 5845(3) Å3, but displayed the same overall topology as 1. The pore contains 18 guest DMSO molecules based on thermogravimetric analysis; however, the crystallographic data suggests a slightly higher value of 22.5 guest DMSO molecules. Similar to ZnCSA·DMF, this was calculated using a combination of the two guest sites crystallographically resolved in the asymmetric unit, giving a refined composition of [Zn7O2(CSA)5DMSO2]·2.64(3)DMSO and the residual electron density calculated by the routine SQUEEZE. This value is reasonably close to upper limit of 27.7 molecules, based on the absolute pore volume (3150 Å3) and a packing density identical to liquid DMSO. The changes in structure from 1 to 3 arise primarily due to the exchange of the two coordinated DMF molecules in the SBU for two DMSO molecules. This process involves a significant rearrangement of the metal cluster; two of the tetrahedral Zn atoms (symmetry equivalents) become octahedral, coordinating to the incoming DMSO molecules and the flexible end of a b axis CSA linker, whereas the two square pyramidal Zn atoms (coordinated to the outgoing DMFs) become tetrahedral, losing their DMF and swapping a connection to the flexible end of a b axis linker for a flexible end of an a axis linker. This change is coupled to conformational adjustments in the flexible ends of the connected a and b axes linkers, which display rotations around the torsions φ, ψ, and δ (values tabulated in Supporting Information S7). In the a axis linkers, these can be seen to change as follows: φ = −151 to 107°, ψ = 60 to −177°, and δ = 173° to 170°. The b axis linker, meanwhile, shows the changes φ = −116 to −73°, ψ = −176 to −174°, and δ = −150 to 167°. In addition, the bend out of plane of the c axis linker is also observed to show a response, with the averaged experimental value changing from 156 to 193°, essentially switching from bending in one direction to bending in the other (see Figure b). A possible mechanism, depicted in Figure a, starts with the binding of the new DMSO molecule to a tetrahedral Zn site without any coordinated solvent (Zn2). This causes a coordinated carboxylate from an a axis linker to rotate, transferring an oxygen (O1) from bonding to Zn2 to bonding to Zn1, the site coordinated to DMF. There is also a shift of a second carboxylate (b axis linker) changing an oxygen (O3) from bonding to Zn1 to bonding to Zn2′ (the symmetry equivalent to Zn2). Finally, the mechanism results in the departure of the original DMF molecule. The carboxylates involved in the cluster reorganization are the central a and b axes linkers connected through the flexible end of the CSA linker; these are ordered and could be accurately characterized by crystallographic techniques, thus enabling the analysis of the coordination change without relying on computational structures. It is worth mentioning that any conformational changes in these linkers during DMF removal were not observed to significantly affect the stability of the a and b directions, thus retaining the pore aperture along the c axis. It should also be noted that the mechanism is mimicked on the opposing side of the SBU with the symmetry generated atoms Zn1′, O1′, O3′, and Zn2′. The large rearrangement of the a axis linker results in a new, much straighter linker conformation, which, combined with the changes to the c axis linker, gives a significant increase in unit cell volume (20% in comparison to 1) and in the void space of the channel running along the crystallographic b axis (see Figure c,d). The solvent accessible void increases to 53.9%, and the pore limiting diameter (now running along the b axis) increases to 11.3 Å. Exchanges with other solvents, MeOH and THF, also showed evidence of structural rearrangements by PXRD (Supporting Figure S7). However, unlike exchange with DMSO, the material did not retain its single crystal crystallinity.

Figure 6

(a) SBU and one linker (situated along the crystallographic a axis) of 1. A proposed mechanism for DMSO exchange to structure 3 is shown. Analogous to Figure , carboxylate carbons colored green are connections running along the a axis, carbons colored purple are connections running along the b axis, and carbons colored orange are connections running along the c axis. Striped colored connections are connections through the flexible end of the linker, solid colors are connections through the rigid end, and half striped connections correspond to a 50/50 distribution through the flexible and rigid ends of the linkers. (b) SBU and one linker (situated along the crystallographic a axis) of 3. (c) The view along the b axis of 1, the Connolly surface for this orientation (calculated using a 1.2 Å probe radius), and the crystallographically modeled a axis linker showing a bent conformation (ψ = 60°). (d) The view along the b axis of 3, the Connolly surface for this orientation (calculated using a 1.2 Å probe radius), and the crystallographically modeled a axis linker showing a straight conformation (ψ = −177°).

In an attempt to experimentally study the behavior of ZnCSA·DMSO during removal of the DMSO contained within the pores but not coordinated to the SBU, a sample was heated in an analogous manner to that of ZnCSA·DMF. The TGA of ZnCSA·DMSO (Supporting Figure S2) showed that the majority of the solvent is removed at 175 °C (higher than the DMF analogue); however, heating at a conservative 150 °C was observed to result in a loss of crystallinity, with the powder pattern (Supporting Figure S8) losing all diffraction peaks except for the strongest peak in the pattern. This remaining peak also showed a reduction in intensity and a significant broadening. It should be noted that in both ZnCSA·DMF and ZnCSA·DMSO, the TGA data suggest that the coordinated solvent is not lost until higher temperatures, roughly 350 °C. The full structural relaxation was therefore instead investigated via DFT optimization, an attempt to predict the behavior, yielding an equilibrium structure (C3) of unit cell parameters equivalent to a = 16.37 Å, b = 17.34 Å, c = 16.60 Å, α = 140.49°, β = 61.60°, γ = 86.61°, and V = 1702.3 Å3. The extent of the overall change, upon removal of only the uncoordinated guests, is predicted to be even larger than for ZnCSA·DMF, the cell of C3 being only 30% of the size of 3. The overall mechanism leads to a large reduction in the accessible void space, now only 2% for C3, and in the pore limiting diameter to 1.2 Å such that any remaining void space can essentially be considered as pockets rather than channels. Because the MOF topology is preserved during the solvent exchange process, in particular, the single and double CSA linker connections, the structural change in ZnCSA·DMSO occurs via qualitatively the same mechanism as in ZnCSA·DMF. Major changes involve out of plane distortions of the single c axis linker through adjustment of the coordination angles θ1 and θ2 and the torsional angles φ, ψ, and δ. However, in ZnCSA·DMSO, this bending of the linker occurs in the opposite direction, relative to its SBU, to that predicted for ZnCSA·DMF (shown in Figure ). This is related to the change in the bending direction of the linker that occurs during the solvent exchange process and means the linker folds over the SBU’s coordinated solvent, which resides on opposite sides in the two materials. The folding in ZnCSA·DMF is also mainly along the crystallographic a axis, whereas the folding in ZnCSA·DMSO is in between the a and b axes, resulting in C3 showing a collapse into the center of the ab plane. This behavior is evidenced from the unit cell changes where ZnCSA·DMF (1 to C1) shows major changes to the β angle (suggesting a mechanism mainly involving the a and c directions), whereas ZnCSA·DMSO (3 to C3) shows a large increase of α (similar to β in ZnCSA·DMF) coupled with a decrease of β (a summary table of unit cell parameters is available in Supporting Information S9). This difference is responsible for the predicted larger reduction in void space, because the collapse of the DMSO structure into the center of the ab plane eliminates the channel running along the original c axis (Figure h) that is still present in ZnCSA·DMF (1) (Figure d). Quantified changes to the single c axis linker, measured between C3 (the optimized structure with fixed experimentally determined cell parameters) and C3, can be found in Supporting Information S8, where they are compared to changes between C1 and C1.

Figure 7

(a) View along the b axis of ZnCSA·DMF (1). CSA linkers colored green are connections running along the a axis, and CSA linkers colored orange are connections running along the c axis. (b) View along the b axis of C1, the predicted DMF structure after guest removal, using the same color scheme. (c) Two SBUs in C1 connected by a single c axis linker. (d) Connolly surface of C1 (calculated using a 1.2 Å probe radius) showing the 1D channel remaining after structure collapse. The channel runs in the direction of the c axis defined for the original structure 1. (e) View along the b axis of ZnCSA·DMSO (3). CSA linkers colored green are connections running along the a axis, and CSA linkers colored orange are connections running along the c axis. (f) View along the b axis of C3, the predicted DMSO structure after guest removal, using the same color scheme. (g) Two SBUs in C3 connected by a single c axis linker. (h) Connolly surface of C3 (calculated using a 1.2 Å probe radius) showing the removal of all channels.

Discussion

The diffraction data collected for ZnCSA reveals that the material exhibits a significant degree of flexibility. This behavior can be divided into two separate responses: A large contraction of the void space during removal of its contained guests and a structural rearrangement caused by the SCCSE of DMF to DMSO. The structural change during guest removal is most obviously seen in the PXRD pattern collected after heating, where the general shift of the Bragg reflections to higher 2θ values is suggestive of a significant reduction in the unit cell dimensions. Indications of the mechanism behind this contraction emerge from the analysis of the single crystal-derived structures of 1 and 2, which show a small scale (Δ9.5% in volume) structural transformation after partial loss of the contained guests. The transformation is largely related to the single c axis linker, with the double linker connections lying along the a and b axes being observed to show very little motion. However, the exact role of the c axis linker in the transformation of 1 to 2 could not be accurately interpreted from crystallographic refinements alone. Unlike the double linker connections along the a and b axes, which are highly ordered throughout the structure sitting antiparallel and related by inversion symmetry, the single c axis linkers coordinate to their SBUs in one of two possible orientations of their rigid and flexible ends. These orientations are randomly distributed throughout the structure and therefore result in high levels of crystallographic disorder, requiring idealized bond distance restraints to generate as a refineable model, which should not be overinterpreted. The interplay between disorder and flexibility, here a consequence of the material topology, is highly intriguing and has been discussed before by Bennet et al.,[68] particularly on the dynamical disorder of UiO-abdc, an analogue of UiO-67 made of azobenzene-4,4′-dicarboxylate linkers.[69]

Periodic DFT calculations, benchmarked by comparing the output to the experimentally determined non-disordered a and b axes linkers, were therefore employed to gain insights into the environment of the changing single c axis linkers. These calculations predict that the transition is a result of a joint mechanism, where the rigid end of the organic linker undergoes changes in carboxylate coordination angle, whereas the flexible aliphatic end shows similar coordination changes combined with conformational adjustments around its sp3 carbons. The coordination angle changes predicted are commonly observed in flexible MOFs built from rigid linkers, particularly wine-rack style materials (e.g., MIL-53[70] or DMOF)[71] and are often referred to as a knee cap or hinge motion.[72,73] Similarly, conformational changes around sp3 carbons are often the key mechanism to many flexible motions reported in frameworks built from solely aliphatic or peptide-based linkers.[16−18] The two types of flexibility appear to lead to very similar distortion to the linker. This is evidenced by the 10° change in ω at both the rigid and flexible ends of the linker, which closely matches the averaged distortion modeled crystallographically. The random distribution of the orientation of this single c axis linker is likely the cause for the similarity, requiring symmetrical responses in order to maintain the more rigid 2D network constructed from the double linker connections along the a and b axes. Importantly, the change to the average bend of the single c axis linker out of the plane of the two SBUs it connects during the transformation of 1 to 2 is reliably captured by the differences between the computationally optimized structures C1 and C2. Therefore, by extending our calculations further, we have also probed the behavior under full removal of the solvent, predicting extended changes on the same joint mechanism, Δω ≈ 50°, resulting in a significantly contracted form with a considerably reduced void space, as suggested by PXRD. The extent of the knee cap motion observed in these calculations is comparable with that of other highly flexible systems. The coordination angle at the rigid end (θ1) is observed to change by 32°, which compares to the 31° response seen in MIL-88-B.[15,73,74] Meanwhile, the torsional changes at the flexible end of the linker observed in ZnCSA are in the range of 100–120°, similar to the rotations observed in peptide-based MOFs displaying structural transformations that are entirely attributed to conformational freedom.[16−18] The CSA linker’s flexibility, expressed through torsional changes, appears to be highly important to the mechanism, because the same flexible behavior was not observed in MOF-123, a framework displaying the same SBU and topology but with a symmetric rigid linker capable of just the knee cap motion. Instead, MOF-123 displays a completely different behavior involving a transition from its non-interpenetrated structure to a twofold interpenetrated structure (MOF-246) upon heating. This occurs at high temperatures (270 °C) capable of removing the coordinated DMF molecules in addition to the guest DMF molecules in the pores, which is outside the scope of the current study on ZnCSA.[22] Furthermore, the previously reported indium-based CSA MOF,[19] displaying a 2D sheet structure built from similar inversion-related antiparallel double linker connections, also does not show the hinge motion observed in ZnCSA. This illustrates the intimate relationship between the linker flexibility and the 3D topology of the MOF (i.e., the presence of both single connections and a flexible linker) in forming this dynamical responsive structure.

The overall response of ZnCSA during removal of its guest species therefore arises from the combined effects of the inherent nature of the CSA linker, which is both asymmetric and flexible, and the topology of the MOF. The double connection of antiparallel oriented asymmetric linkers (i.e., rigid and flexible ends pointing in opposite directions) provides structural stability, impeding distortions in the a and b directions, whereas the randomly oriented single c axis linker remains free and therefore governs the flexible response. The mechanism would be classified as “2D rigid breathing” according to the review by Murdock et al.[75] MOFs displaying this mode tend to consist of rigid 2D sheets made flexible by additional linkers running in the flexible direction.[72,76−82] However, unlike these materials, ZnCSA is built only using one type of linker, and the rigidity in the ab plane is caused by the framework’s topology.

Overview of the responses of ZnCSA. The behavior during DMF to DMSO single crystal coordinated solvent exchange is shown in green. The behavior during uncoordinated guest (DMF or DMSO) removal is shown in yellow.

The framework also shows an interesting structural rearrangement during exchange of its DMF guest species for DMSO. This occurs because exchanging with DMSO not only replaces the solvent located in the pores but also the terminal DMF ligands bound to the SBU. This appears to involve a complicated mechanism where DMSO molecules coordinate to completely different Zn atoms than the original DMF, triggering a rearrangement of the cluster involving changes in the coordination geometry of four out of seven Zn cations and the reorganization of four carboxylate groups. As these carboxylate groups belong to the highly ordered a and b axes linkers, we are able to observe these changes crystallographically. The changing carboxylates are all connected to the flexible end of the linker, suggesting that the solvent exchange can be directly linked to the linker’s inherent flexibility. The process results in changes to the pore size and shape but conserves the topology of the MOF, maintaining the single and double linker connections. Similar single crystal coordinated solvent exchanges (SCCSE) have been applied to a small number of rigid linkers MOFs,[31−36] but SCCSE on MOFs capable of flexible responses has to our knowledge only been reported once by Manos et al.[24] This work describes the structural response of a MOF built from a “semirigid” tricarboxylic linker, one consisting of rigid components connected though a “semirigid” imine (CH=N) linkage that displays limited conformational freedom, during a wide range of topotactic solvent exchanges. However, this framework does not display the large flexibility observed for ZnCSA during removal of its guests, and the flexible response of the framework does not show the cluster rearrangements seen in ZnCSA, which are enabled through the flexibility of the linker.

In order to study the effect this structural rearrangement might then have on the removal of solvent, we returned to using periodic DFT calculations. Although ZnCSA·DMSO showed a qualitatively similar response in the absolute changes of the single c axis linker response, the direction of distortion changed significantly. This is thought to originate from the previous reorganization of the SBU from 1 to 3 upon DMSO coordination and the resulting overall structural changes of the MOF. Consequently, a much more compact structure with a reduced void space is predicted in comparison to the predicted empty structure derived from ZnCSA·DMF. The tuning of the MOF dynamical response by SBU functionalization has been reported for a rigid linker MOF, by Bon et al. for the MOF [Zn3(bpydc)2(HCOO2)] (JLU-Liu4), which exhibits a defined gate pressure response.[36] These workers conducted a systematic substitution of the monocarboxylates in the SBU via solvent exchange, replacing formic acid by acetic acid, benzoic acid, or cinnamic acid, and as a result of the steric effect of the monocarboxylates, the resulting isostructural materials displayed different dynamics during the wine-rack closing. However, this example is for a rigid linker MOF, which, unlike ZnCSA, does not adapt its structure to accommodate the new SBU species. In ZnCSA, the torsions of the flexible ends of the a and b axes linkers actively respond to facilitate the coordinated solvent exchange reaction occurring at the cluster. DMF and DMSO are also comparable in size, whereas the size of the SBU terminal ligand was the major factor in the control exhibited by Bon et al.[36] The predicted changes in the response of ZnCSA·DMSO compared to ZnCSA·DMF therefore illustrate the control possibilities provided by fine-tuning of the SBU functionalization in flexible linker MOFs.

ZnCSA demonstrates the attractive potential of flexible asymmetric linkers, composed of rigid and flexible ends, to control the guest response of MOFs, relying on the delicate balance between topology and flexibility. Further exploration of such linkers, for example, with the extensive Zn-carboxylate cluster chemistry[83−85] as exemplified by the prototypical MOF-5,[86] provides interesting new directions in the design of future flexible MOFs.

Conclusions

We report the synthesis and characterization of a new highly porous flexible MOF constructed from heptanuclear zinc carboxylate secondary building units and flexible asymmetric CSA linkers. Using a combination of crystallographic and computational approaches, we are able to explore the dynamic response of this material during guest removal, which is driven by the changes in the disordered single linker connection along the c axis, while the double antiparallel linker connections along the a and b axes remain locked. The response of the CSA linker reflects its asymmetric nature: its rigid end experiences only a hinge motion, whereas its flexible end shows a combination of a hinge motion and conformational adjustments around its sp3 carbons. The overall anisotropic nature can also be related to the three-dimensional lattice topology within which it is embedded—the same linker in a two-dimensional structure with only one unique linker position does not respond in the same way. Further structural rearrangements of the material are also induced through DMF to DMSO solvent exchange, replacing the two terminal solvent molecules coordinated to the SBU based on a significant rearrangement of the cluster. This occurs through large conformational adjustments to the a and b axes’ CSA linkers, connected to the cluster through their flexible ends, resulting in coordination changes to the individual Zn atoms. The reorganization significantly affects the size and shape of the solvent accessible volume of the material and is predicted to lead to a different structural response during guest removal, which highlights the potential of controlling the dynamic responses of flexible linker MOFs through SBU functionalization.