Unlocking New Topologies in Zr-Based Metal–Organic Frameworks by Combining Linker Flexibility and Building Block Disorder

The outstanding diversity of Zr-based frameworks is inherently linked to the variable coordination geometry of Zr-oxo clusters and the conformational flexibility of the linker, both of which allow for different framework topologies based on the same linker–cluster combination. In addition, intrinsic structural disorder provides a largely unexplored handle to further expand the accessibility of novel metal–organic framework (MOF) structures that can be formed. In this work, we report the concomitant synthesis of three topologically different MOFs based on the same M6O4(OH)4 clusters (M = Zr or Hf) and methane-tetrakis(p-biphenyl-carboxylate) (MTBC) linkers. Two novel structural models are presented based on single-crystal diffraction analysis, namely, cubic c-(4,12)MTBC-M6 and trigonal tr-(4,12)MTBC-M6, which comprise 12-coordinated clusters and 4-coordinated tetrahedral linkers. Notably, the cubic phase features a new architecture based on orientational cluster disorder, which is essential for its formation and has been analyzed by a combination of average structure refinements and diffuse scattering analysis from both powder and single-crystal X-ray diffraction data. The trigonal phase also features structure disorder, although involving both linkers and secondary building units. In both phases, remarkable geometrical distortion of the MTBC linkers illustrates how linker flexibility is also essential for their formation and expands the range of achievable topologies in Zr-based MOFs and its analogues.

* s.canossa@fkf.mpg.de; b.lotsch@fkf.mpg.de  References …………………………………………………..………………………………………………………………... 47 S1. Synthesis of Zr-MTBC phases S1.1. Materials and general information 4',4'',4''',4''''-methanetetrayltetrabiphenyl-4-carboxylic acid (MTBC) was purchased from Alfa Chemistry. Zirconium(IV) chloride (ZrCl4), Hafnium(IV) chloride (HfCl4), benzoic acid, acetone, and N,N-dimethylacetamide (DMA) were purchased from Sigma Aldrich. N,N-diethylformamide (DEF) was purchased from Alfa Aesar. All chemicals were used as received without further purification. Note that the employed ZrCl4 was either stored in the glovebox under inert gas (ZrCl4_GB) or at ambient air (ZrCl4_air) where it was prone for hydrolysis. 1 Ultrasonication was conducted via an ELMASONIC S 100 bath equipped with a high-performance 37 kHz sandwich transducer and state-of-the-art microprocessor. Centrifugation was performed with a benchtop centrifuge Sigma-3-30K from SIGMA. For SEM analysis MOF suspensions were spin-coated onto silicon wafers with a WS-650S-NPP Lite device from Laurell Technology Corporation. S1.2. Synthesis of cubic Zr-MTBC according to literature Zr-MTBC was synthesised according to a procedure published by Ji et al. 2 ZrCl4 (18.2 mg, 0.0781 mmol), MTBC (16 mg, 0.020 mmol), and benzoic acid (554 mg, 4.54 mmol)) were ultrasonically dissolved in DEF (3.2 mL) in a 5 mL microwave vial. The mixture was heated in an Al-block (24 h, pre-heated to 120 °C) under stirring. After cooling to room temperature outside the Al-block, the product was collected by centrifugation and washed with DEF (three times, 16k rpm/15 min/16 °C) and acetone (twice, 16k rpm/15 min/16 °C). The mixture was soaked in acetone over night and washed with acetone once. Supercritical CO2 drying yielded a white powder as product. S1.3. Adapted synthesis of cubic and trigonal (4,12)MTBC-Zr6 and (4,12)MTBC-Zr6 ZrCl4 (18.2 mg, 0.0781 mmol), MTBC (16.0 mg, 0.020 mmol), and benzoic acid (554 mg, 4.54 mmol)) were ultrasonically dissolved in DEF (3.2 mL) in a 20 mL pyrex vial. The mixture was heated in an oven pre-heated to 120 °C for 24 h. After cooling to room temperature outside the Al-block, the formed crystals were washed with DEF twice and acetone three times (soaked in acetone over night during last washing step). The crystals were dried using supercritical CO2. Single crystals of cubic and trigonal (4,12)-MTBC-Hf6 were obtained analogously by reacting HfCl4 (25.0 mg, 0.781 mmol) instead of ZrCl4 for 48 h or 7 d in an oven. S1.4. Screening of reaction conditions in DEF ZrCl4, MTBC, and benzoic acid were ultrasonically dissolved in DEF (3.2 mL) in a 20 mL pyrex vial. The solution was heated in an oven at 100 or 120 °C for 24 h to 7d. In the case of MOF_supernatant, the supernatant obtained after 24 h reaction was further reacted at 120 °C for 72 h. For all reactions, the formed crystals were washed with DEF or DMA twice and acetone three times (soaked in acetone over night during last washing step). The crystals were dried using supercritical CO2.  (2) 120 °C S1.5. Screening of reaction conditions in DMA ZrCl4 (stored at ambient air or new from glovebox), MTBC, benzoic acid, and water (see Table S.1.4.2. for more details) were ultrasonically dissolved in DMA (3.2 mL) in a 20 mL pyrex vial. The solution was heated in an oven (24 h to 12d, pre-heated to 120 °C). For all reactions, the formed crystals were washed with DEF twice and acetone three times (soaked in acetone over night during last washing step). The crystals were dried using supercritical CO2.       S4. X-ray powder diffraction (XRPD) S4.1. Data collection and processing details XRPD patterns were collected at room temperature on a Stoe Stadi-P diffractometer with Cu-Kα1 radiation (λ = 1.540596 Å) or Co-Kα1 radiation (λ = 1.78896 Å), a Ge(111) Johann monochromator, and a DECTRIS Mythen 1K detector in Debye-Scherrer geometry. The samples were loaded into 0.5/0.7 mm inner diameter polyimide capillaries and measured over a range of 2θ = 2.000-30.695˚, with 0.015˚ step size and 50 s counting time per step when using Cu-Kα1 radiation and a range of 2θ = 0.500-115.325˚, with 0.015˚ step size and 200 s counting time per step when using Co-Kα1 radiation. Rietveld refinements were performed with TOPAS v6. 4 Diffraction data collected using Co-Kα1 radiation were used for pure phase cubic refinements. A 2θ offset correction, simple axial model, Lorentzian and Gaussian crystallite size broadening convolutions, and either Gaussian/Lorentzian convolution or Stephens model for strain were used to correct for instrumental and morphological peak-shape effects. The background was described using Chebychev polynomials of 11 th order and One_on_X term for increased low angle background. Scale factors, lattice parameters, and isotropic atomic displacement parameters were refined over a range of 2.5-80° 2θ. Refinements of mixed phase models used data collected from Cu-Kα1 radiation. For these refinements, the background was described with 4 th /5 th order polynomials and One_on_X term. Only scale factors and lattice parameters were refined for the respective phases. A second set of refinements was performed using spherical harmonics corrections of 4 th or 6 th order for majority phases to account for mismatched relative peak intensities.           . Rietveld refinement allowing for reflexes of cubic, trigonal, and tetragonal MTBC-Zr6 against observed PXRD pattern of the product obtained from product from synthesis in DMA for 48 h (MOF_DMA_48h). Simulation 1 includes spherical harmonics corrections of 4-6th order on the predominant phases to account for mismatched relative peak intensities, while simulation 2 does not.  S5. X-ray pair distribution function (XPDF) analysis

S5.1. Data collection and processing details
XPDF analysis was carried out using P02.1, the Powder Diffraction and Total Scattering Beamline, at PETRA III of the Deutsches Elektronen-Synchrotron (DESY). The rapid acquisition PDF method (RAPDF) 5 was used with a large-area 2D PerkinElmer detector (2048×2048 pixels, 200×200 μm 2 each) and sample-to-detector distance of 481.242 mm. The incident energy of the x-rays was 59.795 keV (λ = 0.20735 Å). Samples were loaded into 1 mm inner diameter glass capillaries. An empty capillary was measured as background and subtracted, and a LaB6 standard was measured at room temperature for calibration of the setup. Calibration, polarization correction, and azimuthal integration to 1D diffraction patterns were performed using the software pyFAI. 6,7 Additional total scattering measurements were performed on the same samples in-lab using a Stoe Stadi-P diffractometer with Mo Kα1 radiation (λ = 0.7093 Å), a Ge(111) Johann monochromator, and a DECTRIS Mythen 1K detector in Debye-Scherrer geometry. Measurements were carried out over a range of 2θ = 0.405-110.565˚ and 40.50-110.565˚ with 0.405˚ step size and 150 s count time per step, and 81.00-110.565˚ with 300 s count time. Data ranges were then combined and directly corrected for the 2θ offset of the instrument and polarization. Further correction and normalization of the 1D diffraction intensities were carried out to obtain the total scattering structure function, ( ), which was Fourier transformed to obtain the PDF, ( ) using PDFgetX3 within xPDFsuite. 8,9 The maximum value used in the Fourier transform of the total scattering data was 21.0 Å −1 for the synchrotron and 14.5 Å −1 for the Mo-Kα1 laboratory data. Simulated PDFs were generated from respective models using code from Diffpy-CMI. 10 2). Suitable crystals were isolated and mounted on a MiTeGen loop in a droplet of a perfluoropolyether oil. Diffraction data were acquired using a monochromatic 0.799897 Å radiation at a temperature of 100K, using a cold dry nitrogen stream produced with an Oxford Cryostream. The crystal was placed directly under the Cryostream at 100K without any temperature ramp in going from room temperature to cryogenic conditions. Frames were collected on a Pilatus3S 2M, running a phi scan with oscillation 0.2° and covering an angular range of 360°. Data from c-(4,12)MTBC-Zr6/Hf6 were processed using the software XDS, 12 which included indexing, intensities integration and empirical absorption correction. Data from tr-(4,12)MTBC-Zr6/Hf6 were processed by the software suite CrysalisPro (ver 41.93a), 13 which was used for unit cell determination, intensities integration, and empirical absorption correction. Structure solution and refinement were conducted by using the software suite Olex2 v1.5, 14 using the programs ShelXT (intrinsic phasing) and ShelXL (least squares) respectively. 15 As far as structure modelling for refinement is concerned, a few restraints helped maintaining chemically sound bond distances and molecular geometries during the modelling of structural disorder. Specifically, phenyl rings were constrained to a flat hexagonal geometry by AFIX66, and carboxylate moieties of the linkers were modelled in reference geometries using DFIX and DANG commands. Similarly, ISOR and SIMU ADP constraints were used to guide misbehaving displacement parameters towards more physically sound values, as it was observed that free refinement of displacement ellipsoids led to behaviour inconsistent with a meaningful physically sound situation. Furthermore, hydrogen atoms belonging to OH groups on the clusters or aromatic CH groups were fixed by appropriate ShelX commands and were not refined freely.
Every sample was found as a single crystal, and no twinning was detected in the data or in the reciprocal space maps.
In the final stages of the refinement, the tool "Mask" of Olex2 (alternative to SQUEEZE which is implemented in Platon) was used to account for solvent molecules, assumed as N,N'-dimethylformamide, which were not possible to recognise and model satisfactorily in the residual densities.
Symmetry-independent building blocks were refined with free occupancies to account for missing-building-block defects. When occupancy factors exceeded unity, they were fixed to full occupancy. Additional details for each single crystal are reported below in dedicated sections. Partial linker occupancies in the cubic phases were found to coincide with the presence of unaccounted electron density residues in proximity to metal centres, consistent with the presence of water molecules and OH-groups to ensure overall charge balance of the framework. Therefore, these species were modelled and their occupancy was set to complement those of the linker molecules, so that the sum of the occupancies of linkers and H2O/OH pairs ensures full occupancy on each coordination position around the metal clusters.
Disordered Zr6O4(OH)4 SBUs were modelled without oxo-or hydroxy groups due to excessive overlap of electron densities in the observed data. This created smeared electron density maxima in positions that were not compatible with chemically sound presence of these species. On the other hand, fixing the presence of oxygen and hydroxy groups in idealised positions resulted in strong increase of R1% factors, further signal of inconvenient overfitting. Thus, we opted for not modelling these atoms and instead providing a more clear model of the only metallic component of the clusters.
Difference Fourier maps were computed by using the software Vesta 16 by the following routine: 1) A complete model including structure disorder was refined until convergence, until the structure used for deposition into the Cambridge Structural Database is obtained, which includes the use of solvent mask.
2) The disordered atoms relative to the electron density of interest are removed.
3) A single least square ShelXL refinement cycle is ran, using the command list3 in the ".ins" file, which allows to obtain a ".fcf" file compatible with Vesta.
4) The structure model is loaded in vesta, and the Fourier difference map is generated by the function "Fourier synthesis" and the option "Fo-Fc". 5) Positive residues only are selected in the graphical options, and multi-level isosurfaces are coloured to obtain an informative and clear qualitative picture of the electron density residues due to unmodelled disordered building blocks.

S6.2. SCXRD refinement details
Truncated-octahedral c-(4,12)MTBC-Zr6 crystal nr.1 Linker occupancy: 0.901(3); Ordered cluster occupancy: 1; Disordered cluster occupancy: 0.25 Solvent masking accounted for electron density peaks that were not possible to assign to discrete solvent molecules without deteriorating the quality of the refinements, due to the overlap of symmetry-equivalent positions of solvent species (likely N,N'-dimethylformamide). Difference Fourier maps were computed for displaying the density residues, as shown in Figure S6   Solvent masking produced electron densities without significant differences compared to those shown for crystal nr.1 in Figure S6 (3) The two Zr occupancies were refined to have an overall occupancy of the disordered cluster of 1, since free occupancy refinements resulted in an overall site occupancy >1.
Solvent masking produced electron densities without significant differences compared to those shown in Figure S6  The two Zr occupancies were refined to have an overall occupancy of the disordered cluster of 1, since free occupancy refinements resulted in an overall site occupancy >1. Solvent masking produced electron densities without significant differences compared to those shown in Figure S6.2.2.  Truncated-octahedral c-(4,12)MTBC-Hf6 crystal nr.1 Linker occupancy: 0.885(5); Ordered cluster occupancy: 1; Disordered cluster occupancy: 0.25 Solvent masking produced electron densities without significant differences compared to those shown in Figure S6

tr-(4,12)MTBC-Zr6
The structure is affected by disorder in the form of two alternative framework structures, whose combination is observed in the average structure. Consistently, reciprocal space reconstructions reported in Figure S6.2.10 show streaks and additional intensities along a* and b*, attributable to loss of periodicity in the ab plane of the crystal. To rule out artefacts due to pseudomerohedral twinning or space group symmetry, the structure has been solved and refined in P1, showing the same framework disorder observed in the average structure in P-3c1 settings. An inversion twinning law has been tested on the P1 model, providing refinements where electron densities due to the two alternative framework components were unaffected.

S6.3. 3D electron density maps availability
Electron density maps can be consulted as Vesta files, which are provided as supplementary material. Each deposited structure was used to compute an electron density map by the function "Fourier synthesis…" in the software Vesta using the observed structure factors (Fo). For each structure, two necessary files are provided: a ".vesta" and a ".pgrid" file. Both files need to be present in the same folder, in which case it is possible to open the ".vesta" file using Vesta. At this point, an electron density map will be visible as displayed in the figure S6.3.1. By appropriately selecting data cut-off planes, it is possible to limit the view to specific features of interest, as shown in Figure S6.3.2. It is also possible to change the visualisation of the electron densities by changing the values of the isosurface levels in the dedicated tab (style -> properties -> isosurfaces). The atomic model used to fit the electron densities can be overlayed by simple ticking the "show models" option at the top of the "Styles" tab in the main window.
We highly recommend to consider inspecting such maps when conducting crystallographic modelling, and we encourage to provide these files to allow the scientific community to get access to the crystallographic information obtained from X-ray diffraction experiments, which is the electron densities and not the atomic model introduced to fit them.
To produce these files, one can simply load the .cif file into Vesta, then select "Fourier synthesis…" in the panel "Utilities", click "Import…" and select the .fcf file produced by structure refinements. At this point, it is necessary to select Fo to visualise the observed densities and not the residues or the fitted ones, the latter being Fo-Fc and Fc respectively. Normally, the table is immediately filled with numbers. If it did not, it means that the fcf file was not suited for Vesta. In this case, one should use the LIST 3 command during ShelXL refinements to produce .fcf files compatible with Vesta. If the table is filled correctly, check that there are no non-numerical values associated to some of the reflections, which could cause errors in the calculation of the maps. If present, those reflections should be excluded by selecting and deleting their lines. At this point, clicking "Calculate" will produce the electron density map, which will overlay onto the loaded structure.
This practice can also be used on published structures, provided that the authors included the hkl data (reflections list) in the deposited .cif files as required by the F.A.I.R. principles of scientific research (findability, accessibility, interoperability and reusability of scientific data). Loading a .cif file with embedded reflections in a software for structure solution and refinement such as Olex2 allows to run refinement cycles on the crystallographic model, thereby producing the .fcf file necessary to compute the electron density maps.   The isosurfaces' levels and colors can be changed at will, so that features of interest can be enhanced and better inspected. Note that the only positive fraction of Fo was considered to allow for a cleaner visualization of the map. Normally, inspecting both positive and negative components at the same time allows for a more critical estimate of the real crystallographic information, revealing the presence of artefacts that should not be mistaken for meaningful information.
S7. Single crystal diffuse scattering analysis

S7.1. 3D reciprocal space reconstructions and processing
Reciprocal space reconstructions were generated by the following routine: 1) Each dataset was processed by the software XDS 12 to produce a corrected parameter file (GXPARM) containing unit cell and orientation matrix of the crystal. 2) A suitable mask to cover beamstop and undesired regions of the diffraction frames was created by using the calibration tool of the azimuthal integration library pyFAI (pyFAI-calib2), 6 and used for all datasets collected with the same instrumental configuration and settings. 3) Starting, unedited reconstructions were produced by the software Meerkat 17 by setting as maximum h, k, and l 58.5, and number of pixels 1171 to achieve a sampling step of 0.1 reciprocal lattice units. 4) A series of corrections were adopted by using a custom-made python program: a) Symmetry averaging has been applied on each reconstruction to improve statistics, decrease noise, and cover missing volumes, using the Laue symmetry of the average structure model (m-3m). b) A spherically symmetric background was extracted from the reconstructed diffraction pattern and then subtracted from the symmetry-averaged reconstruction. The background was estimated by calculating, for each Q value of the reconstructed reciprocal space, the mean value of the 1% weakest intensities observed at that Q-value. The choice of the 1% threshold was obtained by comparing the background subtracted patterns obtained using different threshold values, such as 0.5%, 1%, 2%, 5%. The judged the 1% threshold value as sufficient for eliminating most of the visible amorphous background. As no quantitative usage of the diffuse intensities was intended, we did not proceed further with background subtraction optimization. We note, in passing, that this procedure provides a semi-quantitative selfconsistent method for the correction of extrinsic as well as Compton scattering contributions to the total scattering. c) To prevent artefacts in the 3D PDF function, the intensities below 2.5 and above 58 reciprocal lattice units along each of the main axis were set to zero. This procedure produces a spherical reconstructed reciprocal space. d) Each reconstruction was normalized by dividing the value in each voxel by the total integral of the reconstruction. A multiplier (10 12 ) was then applied post-normalization to each reconstruction.
Bragg peaks removal was based on a "punch-and-fill" routine. 18 Since the diffuse intensities are mostly present around the Bragg positions, a spherical mask with fixed radius was applied to a voxel around each Bragg position (effectively "punch" the Bragg intensities). The new intensities on the masked voxels were calculated by interpolating 19 the intensities within a spherical volume centred on the same Bragg position but with a radius twice the mask radius.
3D PDF maps were calculated by applying a Fast Fourier transform (FFT) to the corrected reciprocal space reconstructions, while 3DΔPDF maps were calculated by the same procedure, but on the Bragg-subtracted reconstructions after the punch-and-fill procedure. No further normalization was conducted after Bragg peaks removal or on the PDF maps.            The absence of Zr-Zr correlations is evident from noticing that any dashed line crosses only one Zr atom. Therefore, for any given plane parallel to the main planes (ab, ac or bc) there is only one Zr atom belonging to a single cluster.

S7.2. Diffuse scattering simulations
To simulate the diffuse scattering patterns originating from a disordered c-(4,12)MTBC-Zr6 crystals with random cluster disorder, we used the program Zürich Oak-Ridge Disorder Simulation (ZODS). 20 As average structure model, we used the structure refined from the truncated-octahedral crystals, and kept all atomic positions fixed in their average sites. In the disordered sites, 4 mutually exclusive chemical species were defined, corresponding to a Zr6O4(OH)4 cluster oriented in one of the four possible alternatives. As we did not define any correlation parameters among those sites, in each site the selection of the cluster orientation to be used was completely random, resulting in what we refer to as "random cluster disorder". Since no refinement of computed diffuse pattern against experimental one was conducted, we used only an arbitrary number of 100 MC cycles, and simulated 10 single crystal of 20x20x3 unit cells.
For the intensity calculation, we used the Fourier section of the software suite Discus (ver. 6.07.00) and instructed the program to subtract the average structure factor to eliminate Bragg reflections from the computed patterns. Intensities were calculated for the only hk0 plane, from -58.5 to +58.5 of h and k, 1171 steps for each reciprocal space direction, and using 10 lots of 10x10 unit cells for each of the crystals generated by ZODS. In this way, 10 calculated hk0 reciprocal space slabs were obtained, and averaged to produce the final pattern shown in the manuscript.
The latter correspond to the models shown on the right side of Figure 6 in violet and turquoise. The underlying nets 26 are defined for octahedral M6 SBU with tetrahedral ligand tet, MTBC and correspond to a general formula (SBU)(tet)n. The SBU assume three different geometries: cubic cub, icosahedral ico or cuboctahedral cuo. We use the three letter symbols from polyhedra and nets from RCSR (see http://rcsr.net/ ). 27

S9. Remarks on the multiphase behaviour in Zr-MTBC MOFs
Although Zr6O4(OH)4 cluster based MOFs have been intensively studied since 2008, the control over MOF phases remains challenging. Oftentimes, multiple MOF topologies with different catalytic activities and gas sorption properties can be built from the same linker and Zr-cluster. [21][22][23][24] We identified three MOF phases, cubic, tetragonal, and trigonal Zr-MTBC to form from ZrCl4 and MTBC when using benzoic acid as modulator. The phase composition of the product was found to vary depending on the reaction conditions ( Figure S5.1). Changing the reaction time, temperature, concentration of reactant, solvent, and water content during synthesis from ZrCl4 led to several observations worth mentioning, which we list in the following.
(1) Both reaction temperature as well as concentrations of linker and ZrCl4 did not significantly affected the composition of MOF phases formed, all yielding cubic Zr-MTBC as main product. As discussed above, decreasing both MTBC and ZrCl4 concentration, however, remarkably slowed down particle growth (24 h vs 5 d) and yielded c-(4,12)MTBC-Zr6 crystals with higher crystallinity and a cube-shaped crystal habit instead of a truncated-octahedral one.
(2) Increasing the reaction time, in turn, inversed the phase ratios in the product and more tetragonal and trigonal Zr-MTBC were formed.
(3) Importantly, separation of the supernatant after 24 h reaction and further reaction of the supernatant mostly yielded the tetragonal phases. This suggests, that at the beginning of the reaction cubic and trigonal Zr-MTBC form while tetragonal Zr-MTBC growth later in the reaction.
(4) When employing DMA as solvent instead of DEF, no cubic Zr-MTBC was formed. Similar to the reactions in DEF, longer reaction times favored the formation of tetragonal Zr-MTBC when using DMA as solvent. For instance, reaction from ZrCl4_air yielded a mixture of the trigonal and tetragonal phase after 24 h whereas reaction for 48 h mainly yielded the tetragonal MOF suggesting a phase transformation from the trigonal to tetragonal phase.
To better control reaction kinetics, synthesis of Zr-MTBC MOFs in DMA was further performed using neat ZrCl4_GB under addition of small amounts of water, based on our recent publication on the effect of water during synthesis of porphyritic Zr-MOFs. 1 Microscope images revealed that the amount of water added to the reaction mixture affected the phase composition of the products. Reaction under addition of 4 µL water yielded hexagonal particles whilst synthesis with 8 or 12 µL water yielded mixtures of hexagonal and octahedral particles, characteristic for trigonal and tetragonal Zr-MTBC, respectively. Reaction time affects the phase composition of the product: synthesis with 4 µL water yielded hexagonal particles after 1 d (MOF_DMA_1d_4) and 2 d (MOF_DMA_2d_4) reaction whereas longer reaction times increased the amount of octahedral particles in the product.
Regrettably, PXRD measurements of the products containing hexagonal particles only had poor crystallinity (see Section S4.3), and an unambiguous assignment of peaks to the trigonal phase is not possible. In the absence of phase homogeneity information from PXRD, our conclusions can only be based on the morphological characteristics of the product as hexagonal platelets. Although this morphology has only been observed for the trigonal phase we report, we cannot claim that this is not an additional polymorph of Zr-MTBC.

S10. Nitrogen sorption analysis
Sorption measurements were acquired on a Quantachrome Instruments Autosorb iQ 3 with nitrogen at 77 K. Samples were activated under high vacuum at 120 °C for 12 h before measurement unless stated otherwise. The pore size distribution (PSD) was determined from nitrogen adsorption isotherms using the QSDFT (cylindrical pores, adsorption branch) kernel in carbon for nitrogen at 77 K implemented in the ASiQwin software v 3.01.