High-Throughput Microscale MOF Catalysis via Solid Dispersion Reactor Dosing

Abstract

Metal–organic frameworks (MOFs) exhibit exceptional structural diversity and tunable chemical environments, offering excellent opportunities for tuning the selectivity of catalytic centers confined to their pores. As such, MOF catalysts have been shown to exhibit distinct size-, regio-, and chemoselectivity relative to other heterogeneous platforms. Despite these advantages, the discovery and optimization of highly active and selective MOF catalysts for complex transformations remain challenging. Their poor solubility, low bulk density, and mechanical fragility complicate their integration into small-scale, high-throughput (HT) experiments that would otherwise facilitate catalyst optimization. In this work, we present the adaptation of inexpensive bead-based reactor dosing methodologies to overcome these limitations. A central feature of this strategy is the development of plate-reader–based analysis of reactor dosing, which enables the rapid optimization of MOF–bead formulations for accurate and precise well-to-well dosing. The utility of this approach is demonstrated in a proof-of-concept study involving the identification and optimization of a novel MOF-catalyzed C─H oxidation reaction, thereby establishing an accessible workflow for accelerating MOF catalyst development in batch reaction formats.

Graphical Abstract

1. INTRODUCTION

Metal–organic frameworks (MOFs) are a highly versatile class of crystalline materials, distinguished from other porous materials by the combination of high surface areas, topological diversity, and extensive structural tunability.^1,2 Constructed from inorganic nodes coordinated to organic linkers, the diversity of their component parts enables the rational tailoring of pore sizes, shapes, and functionalities. This immense tunability provides opportunities for manipulating interactions with guest molecules.^3-5 Indeed, there are numerous examples of MOF-based catalysts, adsorbents, and sensors where the internal structure and functionality of the MOF has been precisely tuned to realize high selectivity for their application.^6-14

The high tunability of MOF catalyst active sites and microenvironments is particularly attractive for catalytic contexts where traditional manipulation of the immediate catalyst active site is insufficient to effect high selectivity. For instance, Long et al. reported that the pore size and hydrophobicity of Fe₂(dobdc) derivatives can be altered to affect ketone-alcohol selectivity of a C─H oxidation.¹⁰ As another example, Ma and coworkers described the modification of a MOF-based hydrogenation catalyst’s secondary sphere to effect enantioselective hydrogenation.⁸ These reports highlight the potential of MOFs in selectivity-limited transformations but also underscore one of the key challenges in this area: identification and optimization of an appropriate MOF catalyst. Minor adjustments to metal centers, linkers, pore dimensions, or defect densities can significantly alter reaction outcomes but the impact of changes to these structural features is often hard to predict.^15-17 Moreover, though elaborate tuning of pore functionality is possible with MOFs, scalable syntheses of highly engineered MOFs is often difficult. Significant material demands of subsequent optimization campaigns is often limiting. These problems are not easily overcome with machine-learning based approaches to focus synthetic efforts. Such approaches are generally of limited utility without extensive training sets.^18,19 As such, exploitation of MOF catalysts’ tremendous synthetic tunability in catalysis is hampered by subsequent labor intensive screening of numerous candidate materials and reaction conditions.

High-throughput experimentation is an effective strategy for identifying optimal conditions with limited quantities of catalyst. ^20-22 These approaches have transformed other areas of catalysis such as biocatalysis and transition metal catalysis by accelerating catalyst optimization and minimizing the quantity of material required.²³ Yet, application of these powerful screening methodologies remain rare in the context of MOF catalysis. Just a handful of examples using advanced reactors and powder handlers have been reported, highlighting the potential for such technologies and need for more accessible approaches.^24,25

MOFs are generally insoluble, limiting straightforward use of automated liquid handlers that dominate most high-throughput catalysis workflows. Moreover, relative to other classes of heterogeneous catalysts, MOFs suffer from limited mechanical stability, a high susceptibility to static electricity accumulation, and low density.²⁶ These properties generally complicate their accurate dosing into reactor wells on small scales (<1 mg). Modern powder dispensing robots have addressed some of these issues in the context of MOF catalysis but are currently expensive, limiting their widespread adoption.^25,27 Promising studies of slurry based approaches have been described but the generality of these strategies across different MOF classes in additional catalytic contexts is not yet established.²⁸

The difficulty in high-throughput handling of insoluble solids is a long-standing challenge in high-throughput chemistries.^29,30 One effective strategy to address this problem without expensive powder handling robots is to improve the processibility of the target molecule by adsorbing it to the surface of the bead. The dense, typically spherical, beads improve the flowability of the powder and dilutes the adsorbate, functioning as a solid-state equivalent of a stock solution. This approach enables efficient volumetric dosing into microscale reactors with either scoops or powder-handling devices. Recent developments by researchers at AbbVie as well as academic groups have formalized these strategies and demonstrated their practicality even without advanced instrumentation.^31-33 Known as ChemBeads, these inorganic or polymeric beads enable efficient small-scale screening of extensive molecular libraries by improving reproducible dosing without the use of binding agents, which have potential to impact the properties of the adsorbate. With appropriate preparation of the ChemBeads, the target molecule can be precisely and rapidly dosed into the reactor using calibrated scoops. This simple yet powerful technology has made high-throughput catalyst screens accessible to labs without advanced robotics but is not widely applied to MOF catalysis. MOF are significantly different than prototypical adsorbates evaluated in ChemBeads, which are typically mixed with dense inorganics or organics that can be impacted on to the surface of the bead. MOF are typically low-density solids with widely varying morphologies and varied mechanical stabilities. These properties make it unlikely that exact translation of existing ChemBead approaches will be universally applicable to MOFs and may therefore require targeting fundamentally different interactions with the beads.

In this work, we demonstrate a methodology that adapts bead-based methods to achieve consistent, small-scale dosing of MOFs for microscale reactions (Figure 1). We show that conventional bead formulation approaches developed for molecular ChemBeads do not work well for MOFs, but that formulations specifically optimized for MOFs can result in satisfactory microscale reactor dosing. A significant challenge in MOF-bead dispersion (MOFBead) optimization is the rapid and economical characterization of reactor dosing precision and accuracy. The variability in reactor dosing, a critical factor affecting the utility of the system, can only be established through large numbers of individual trials, which are labor intensive to efficiently characterize with MOFs. To address this, we first describe the development and benchmarking of a plate reader-based approach for quantifying the variability in reactor dosing with MOFBeads. We then apply this analytical approach to the optimization of MOF-glass bead dispersions that result in a consistent, general, and lab-accessible method for precise dosing of MOFs into reactor wells at scales down to 100 μg of MOF per reactor. Finally, this method is then applied to the identification and optimization of a novel MOF-based C─H oxidation catalyst in a parallel screen of over 150 catalyst and condition combinations using less than 50 mg of MOF. This work therefore establishes an accessible approach for higher throughput workflows in MOF catalysis.

Figure 1.

Overview of this work. Dispersion of MOFs in glass beads (MOFBeads) to enable rapid, small-scale dosing of MOFs into reactors.

2. METHODS

General:

All chemicals were used as received unless otherwise stated. All manipulations were done under ambient atmosphere unless otherwise noted. Ni-MOF-74, UiO-67-bpy, NU-1000, HKUST-1, MOF-808, NiBDP, and MFU-4l were prepared by following or modifying from the literature procedures. ^30-39 Multichannel pipette was purchased from Eppendorf (Research^® Plus Move It^®, 6-channel, 120-1200 μL); 24-, 48-, and 96-well Para-Dox aluminum reactor blocks were purchased from Analytical Sales and Services, Inc.; Agilent 8453 UV-Vis was used for measuring λ_max of the digested MOFs; 3D-printed scoops were donated from AbbVie Inc.; Agilent Cytation 1-CYT1FA Multimode Microplate Reader (100 to 1000 cm⁻¹) was used for high-throughput characterization of reactor dosing; 96-Well Microtest Plate (Polystyrene, flat-bottom) was purchased from Universal Medical and used for plate reader without lid.

Validation of Plate Reader Based Characterization of Reactor Dosing:

The UV-Vis plate reader absorbance assay for characterizing the quantity and variability of MOF delivered to each reactor well was validated by comparison against X-ray fluorescence spectroscopy. Calibration curves were made by digesting known quantities of the parent MOF, as opposed to making the calibration curve from standard solutions of the metal (for XRF) or linker (for plate reader analysis). By making the calibration curves from the native material, corrections for defects are not necessary. New calibration curves were made for each new optimization using the parent MOF as some variability was noted day-to-day.

The standard digestion calibration curves were obtained by weighing 10.0 mg of Ni-MOF-74 in a 20 mL glass vial. This MOF was then digested with 10 mL of H₂SO₄ in DMSO (1.86 M) by stirring at 500 rpm using a 1 cm PTFE magnetic stir bar at 50 °C for 16 h. The 24 μL of this digested solution was then pipetted into a 96-Well Microtest Plate and diluted by adding 216 μL DMSO to it to obtain the 0.1 mg/mL initial concentration for serials dilution, which was conducted 4 times on the same plate with the dilution factor 2 (0.05 mg/mL, 0.025 mg/mL, 0.0125 mg/mL, and 0.0063 mg/mL). To add additional concentrations in between of each point, the second set of serial dilution was performed on the same plate with 0.075 mg/mL starting concentration by pipetting 18 μL of the same parent digested solution to 222 μL of DMSO, and the serials dilution was performed 4 times from it with dilution factor 2 (0.0375 mg/mL, 0.0188 mg/mL, 0.0094 mg/mL, and 0.0047 mg/mL). Each set of serial dilutions was performed twice to obtain the average. These solutions were then characterized by the plate reader at 372 cm⁻¹, and the absorbance of each concentration was recorded. The standard digestion calibration curve was then plotted in absorbance against concentration. Similar calibration curve of MOF concentration vs fluorescence intensity is plotted for XRF.

To compare the two calibration methods for accuracy, 24 MOFBead reactor dosing trials were analyzed by both plate reader and XRF. In a representative trial, the plate reader samples had a coefficient of variation of 6.1% with an average yield of 79.5% relative to the theoretical concentration of linker, while XRF had and average yield of 78.2% relative to the theoretical concentration of Ni in the solutions with a coefficient of variation of 15.8%. The plate reader was used for subsequent dispersion optimizations on account of the lower cost and faster data acquisition.

General Procedure for Characterization of Volumetric MOFBeads Reactor Dosing Accuracy and Precision:

Reactor dosing accuracy and variability were characterized using an Agilent Cytation 1-CYT1FA Multimode Microplate Reader with a measuring wavelength of 100 to 1000 cm⁻¹. The MOF-Bead dispersions were carefully scooped into 24 reactor wells (2 mL glass vial inserts, Para-dox Standard 48-position Parallel Synthesis Reaction Block, aluminum, Analytical Sales and Services, Inc) using calibrated volumetric scoops. For all MOFs but MFU-4l derivatives, digestion was conducted with 1 mL of 10% v/v H₂SO₄ in DMSO (1.86 M). This solution was pipetted into the reactor wells by the multichannel pipette (Eppendorf Research^® plus Move it^®, 6-channel, 120-1200 μL). For MFU-4l, the digestion was conducted by dissolving the material in 1 mL of DMSO containing 10% by volume NBu₄OH (40% v/v in H₂O). The mixtures were then heated at 50 °C in the reactor block for 16 h at which point 12 μL of the solution were taken from the digestion vials and transferred to a 96-Well Microtest Plate (Polystyrene, flat-bottom, Universal Medical) using the multichannel pipette. Each well was further diluted with 108 μL of DMSO. The absorbance of each well was then measured at the λ_max for each MOF. The concentration of the MOF in each well was then determined by comparison to the calibration curves of each material.

Characterization of MOFBeads by XRF:

XRF was used to validate the plate reader characterization. The Shimadzu EDX-7000 Energy Dispersive X-Ray Fluorescence Spectrometer was used for this measurement, using the following parameters (Emission current: 100 μA; High Voltage: 50kV; Atmosphere: Air). Samples were prepared in a 33 mm diameter plastic XRF sample cup with polypropylene TF-260-3410 thin film. XRF samples were prepared by digesting MOFBeads in acid or base as in plate reader samples above. The digested solutions are directly pipetted into XRF cups and analyzed.

Characterization of MOFBeads by PXRD:

The structural integrity of the MOFBead dispersions was evaluated by PXRD following extraction from the beads. The relatively low sedimentation rate of the MOF component relative to the glass component enabled the separation by decanting an agitated solution of the MOFBead via the following procedure: in a 15 mL polypropylene centrifuge tube, the approximately 200 mg of MOF-Bead dispersions were manually shaken in 5 mL of acetone for ~5s. The suspension was allowed to settle for a 5 s at which point the supernatant was carefully decanted, leaving the glass beads at the bottom. Fresh acetone (5 mL) was added to the beads, and this process was repeated three times to recover residual MOF trapped on the beads. The combined acetone supernatants from the washes were then removed under a gentle stream of N₂ at room temperature until dry. Powder X-ray diffraction patterns were collected on this material which confirmed that the crystallinity was maintained.

Characterization of MOFBeads by N₂ Adsorption Porosimetry:

The porosity of the MOFBead dispersions was evaluated by N₂ gas adsorption isotherms. Approximately 1000 mg of MOFBeads (5 wt% MOF) were loaded into 1 mL glass vials which were pre-tared together with gas adsorption glass analysis tubes (Micromeritics). These vials were then inserted into the analysis tubes and carefully slid to the bottom, and the tubes were capped with Transeals or CheckSeals. The MOFBead samples in the sample tubes were activated for 3 to 5 hours on a Micromeritics VacPrep 061 (~3×10⁻² mmHg). Nitrogen adsorption isotherms were obtained using ultra-high purity nitrogen in a 77 K liquid-N₂ bath. The activation temperatures matched those used to make the parent MOF before dispersion in glass beads. It should be noted that the reported surface areas assume exactly a 5.0% mass fraction of the MOF. Thus, the actual surface areas are almost certainly higher than reported because our analytical data show that MOF loading is generally ~3.7–4.8%, depending on the MOF. However, because the isotherm sample tubes were loaded differently (not with scoops) than the reactors from which this data was derived, the surface areas were not corrected by the reactor dosing mass fraction data. The surface area of the pure glass beads was measured independently and found to contribute less than 5 m² g⁻¹ to each sample.

General Procedure for the Preparation of MOFBeads:

To ensure a uniform mixture is achieved, the as-synthesized MOF, typically 20-100 mg, is first gently ground using an agate mortar and pestle into a uniform powder without visible clumps. This MOF was then recovered from the mortar, transferred to a tared 20 mL scintillation vial using laboratory weighing paper, then diluted with the appropriate quantity of glass beads, based on the amount transferred (not the amount initially added to the mortar and pestle) to reach the target weight %. The target concentrations were typically 1.5, 2.5, or 5.0 wt%. A 1 cm PTFE magnetic stir bar was then added, stirred at 200 rpm for 1~2 min, then removed from the stir plate to obtain a uniform mixture.

Small Scale Synthesis of M-MFU-4l-X Derivatives:

Cation and anion exchanged MFU-4l derivatives were synthesized in parallel using a modified procedure of Volkmer et al.⁴⁰ Mn, Ni, Co and Cu, stock solutions of the metal salts were prepared by dissolving each of them in 20 mL anhydrous DMA (5 mmol, which were 990 mg of MnCl₂•4H₂O, 1207 mg of NiCl₂•6H₂O, 660 mg of CoCl₂ and 682 mg of CuCl₂). In a 24-well aluminum reaction block (with 4 mL vials insert), each vial was charged with a 3 × 5 mm PTFE magnetic stir bar and 12.5 mg (0.01 mmol) of the parent Zn-MFU-4l-Cl. Then 2 mL of the metal stock solutions (0.5 mmol) were pipetted into the vials according to rows (row A: CoCl₂, row B: NiCl₂ •6H₂O, row C: CuCl₂, row D: MnCl₂•4H₂O). The reaction block was then sealed using the aluminum block cover containing a rubber mat and a PTFE sheet layer. The reaction mixtures were stirred at 500 rpm at 60 °C for 20 hours. The MOFs were then pelletized by centrifuging the reactor block in a Thermo Scientific Sorvall ST8 Centrifuge equipped with M-10 Microplate Swinging Bucket Rotor (at 900 rpm, Relative Centrifugal Force (RCF) = 108) for 20 minutes. The supernatants were carefully removed by pipette and washed three times of each solvent with DMA, MeOH, and DCM. This washing process could be achieved in parallel by centrifuging the whole reaction block containing the MOF derivatives using the above conditions, then carefully pipetting out most of the supernatants in each vial, care was taken not to pipette out MOF. The washing step was done using an Opentrons OT-2 liquid handling robot. In this case, fresh wash solvents (1.5 mL) were added to each vial using the robot, and the reaction block were sealed and gently agitated to mix the MOF with the fresh solvent, the resulting mixtures were then allowed to stir gently at 300 rpm for 10 minutes. The block was then centrifugated on a plate centrifuge rotor, and the supernatants were pipetted out and replaced with fresh solvents by the OT-2. This cycle was repeated three times for each solvent. After the washing process, the whole block containing resultant M-MFU-4l derivatives were directly put in a vacuum oven and the MOFs were activated at 150 °C for 18 h.

A similar process was applied to achieve the anion ligand exchange, which were OTf, BF₄, OAc, Br, and NO₃ anions. In this case, 3 mL of 0.2M solutions of the lithium salts (LiOTf, LiBF₄, LiOAc, LiBr, LiNO₃) in MeOH were added to the previously obtained M-MFU-4l derivatives. The reaction block was sealed and stirred at room temperature for 30 min. The resultant M-MFU-4l-X derivatives were pelleted by centrifuge and washed three times with each solvent of MeOH and DCM (2 mL) using the pipetting robot.

Catalytic Screening of M-MFU-4l-X Derivatives:

The 24 M-MFU-4l-X (M= Mn, Co, Ni, Cu; X= Cl, Br, OAc, NO₃, BF₄, OTf) MOFs were mixed with glass beads for obtaining MOF-Beads and screened for catalyzing 1-bromo-4-n-butylbenzene oxidation against 4 oxidants (PIDA, TBHP, H₂O and Air), making a total of 100 conditions once control reactions (MOF-free) are added. This reaction was set-up using 1-bromo-4-n-butylbenzene (0.00825 mmol, 1 eq.) as the limiting agent, with 3.5 eq. of oxidants (0.0289 mmol) and scoops of the M-MFU-4l-X MOFBeads.

In a 96-well aluminum reaction block, 48 of the 1 mL glass vials, alternating with one empty space between, were charged with 3 × 5 mm PTFE magnetic stir bars. The 24 different M-MFU-4l-X MOFs were scooped into each vial as MOFBeads (pure glass beads were scooped into the 4 control vials), and 1-bromo-4-n-butylbenzene (34.5 μL of the 1/23 v/v stock solution in MeCN, 0.00825 mmol, 1 eq.) were pipetted into each vial as stock solutions. Next, 120 μL of fresh MeCN was added to all the vials, followed by the addition of the oxidants via weighing or pipetting (9.3 mg PIDA, 4.0 μL TBHP, or 2.95 μL H₂O₂ aqueous solution, 30 wt%). The reaction block was then sealed with an aluminum cover containing a rubber mat and PTFE sheet layer in between, and heated to 65 °C. After stirring at 300 rpm for 18 h, the reaction block was cooled by placing an ice bath underneath, and a hot sand bath on the top for about half an hour for generating the temperature gradient to the block, in order to collect the liquids that were vaporized and condensed on the PTFE sheet. The aluminum cover, rubber mat and PTFE sheet was then unscrewed and removed, and the reactions were diluted with CDCl₃ using a multichannel pipette by adding 587 μL of internal standard stock solution, which trimethyl-tricarboxylate benzene was pre-dissolved in it in a concentration of 3.54 mg/mL (0.014 mmol/mL). The reaction block was resealed and stirred at room temperature for 2 h, before centrifuging the plate to precipitate the solids (300 rpm, RCF=10). The clear solution on top was then pipetted into NMR tubes and reaction yield was quantified by NMR.

3. RESULTS AND DISCUSSION

Plate Reader Quantitation of Reactor Dosing

Due to the challenging handling characteristics and limited mechanical stability of many MOF materials, we anticipated that the formulation of MOF–bead dispersions would require different conditions and characterization than formulations involving molecular compounds or other classes of insoluble solids. As such, we first focused on developing a rapid, general, and high-throughput method to assess reactor dosing accuracy and variability tailored specifically for MOFs. We hypothesized that dosing precision could be evaluated through digestion followed by plate-reader analysis to quantify the initial amount of MOF delivered to each reactor well. Importantly, UV-vis quantitation has been used to characterize surface functionalization of MOFs with chromophores.⁴¹ Given that most metal–organic frameworks incorporate UV-active linkers, and most are easily digested using strong acid or base, we reasoned that plate reader analysis of reactor dosing would be general across many classes of MOFs.

To test plate reader quantitation of reactor dosing, a known mass of a prototypical MOF, Ni-MOF-74, was digested in strong acid, serially diluted, and analyzed by UV–vis spectroscopy using a plate reader. The resulting calibration curve was linear over two orders of magnitude in concentration, confirming that the linker absorption band provides a robust quantitative measure of reactor dosing. To further validate this approach, these samples were independently analyzed by X-ray fluorescence (XRF) following digestion. The two methods showed excellent agreement with each other (Page S18). Importantly, the plate-reader method offers a substantial advantage in throughput over other methods such as NMR and or even XRF, enabling parallel quantification of up to 96 reactor dosing trials within seconds, compared to approximately two minutes per dosing trial for XRF or NMR. This efficiency makes the method particularly well suited for systematic optimization of MOF–bead formulation protocols, where repeated dosing trials are required to accurately establish variability statistics.

Optimization of MOFBead Dispersions

Having established a reliable high-throughput method for assessing small quantities of MOFs dosed into reactors, we next evaluated the consistency of volumetric bead-based dosing for MOFs. Ni-MOF-74 was again chosen as a model material due to its ease of preparation and numerous applications. Though we expected that some modifications to established ChemBead formulation protocols would be necessary, we chose unfunctionalized 213-300 μm glass beads as a starting point based on their compatibility with molecular compound ChemBead formulations in previous reports.³³

Dispersions of Ni-MOF-74 in glass beads were prepared using vortexing, mechanical stirring, and solvent-assisted methods for mixing (see page S7 for details). Resonant acoustic mixing, while highly effective in molecular ChemBeads, was not investigated due to limited instrument accessibility. Unfortunately, none of these mixing methods produced uniform dispersions. Either visible MOF aggregates persisted, or excessive static charge led to MOF adhesion to vial walls, both of which precluded reproducible dosing. In contrast, manual grinding with a mortar and pestle followed by stirring with the glass beads yielded visually homogeneous mixtures with minimal static electricity accumulation (Figure 2a).³² However, volumetric dosing of the Ni-MOF-74 dispersion in 213–300 μm glass beads did not yield uniform reactor loading. Analysis of 24 dosing trials using volumetric scoops by plate reader yielded a coefficient of variation in well-to-well dosing of 35.9%. Reactor dosing accuracy was also poor. The apparent MOF mass fraction of the beads was ~0.8 wt%, compared to the expected 5.0 wt% based on the component masses. Although some material loss to the storage container and transfer tools is anticipated, the observed accuracy and precision was far too low to be practical.

Figure 2.

(a) Overview of dispersion process and plate reader characterization of reactor dosing (b) boxplot showing the impact of bead size on reactor dosing accuracy and precision (c) boxplot showing the influence of scoop size on reactor dosing variability and precision using < 106 μm beads and (d) boxplot showing how MOF mass loading in < 106 μm beads influence reactor dosing accuracy and precision.

We next investigated the influence of bead composition on reactor dosing consistency. We hypothesized that less polar surfaces may result in better MOF adhesion. Accordingly, Ni-MOF-74 was dispersed at 5.0 wt% in various bead types including silanized glass beads (213–300 μm), polystyrene beads (354–500 μm), and polytetrafluoroethylene beads (PTFE, 350 μm). Composition exerted a pronounced effect on performance. Silanized glass and PTFE beads exhibited excessive static accumulation, leading to visibly heterogeneous mixtures and significant adhesion to the vessel walls. Variation in the preparation vessel had negligible impact on reducing static accumulations and these formulations were therefore not pursued further. In contrast, the polystyrene bead formulation produced visually homogeneous, free-flowing dispersions (Figure S6). However, polystyrene beads swelled excessively upon organic solvent exposure, a property reasonably anticipated to complicate their use in non-aqueous batch catalysis. Consequently, polystyrene beads were also excluded in favor of the unfunctionalized glass beads.

Having identified glass beads as the most viable solid dispersant, we next sought to enhance reactor dosing consistency by optimizing bead size. Parallel dosing experiments (n=24) demonstrated that larger glass beads (425–600 μm) resulted in comparably accurate but even less precise dosing than initial trials with 213–300 μm beads (Figure 2b). Again, significant adhesion of MOF to other glass surfaces and the plastic scoop was noted. These dispersions exhibited similarly low apparent MOF loadings of approximately ~1.0 wt%, compared to the expected 5.0 wt%. Moreover, dosing variability increased, with a standard deviation of 66.5%. In contrast, reducing the bead size to glass beads below 106 μm markedly improved both dosing accuracy and precision. Across 24 dosing trials, 73.3% of the expected MOF mass was delivered to each reactor with a standard deviation of 8.7%. This degree of variation is comparable to accepted tolerances in other bead-based high-throughput dosing.^32,33

Both the mass fraction of MOF in the dispersion and the size of the volumetric scoop is reasonably expected to impact how much MOF is delivered to each reactor. As such, we examined the effects of MOF loading within the beads and the scoop size used for reactor dosing. As shown in Figure 2c, smaller scoops performed comparably to larger ones, with no significant change in dosing variability (Table S4). Notably, the 2 mg scoop containing 5 wt% MOF, which delivers approximately 100 μg of MOF per reactor, achieved a coefficient of variation below 11% across 24 trials and good absolute MOF concentration of ~3.9 wt%.

Dispersion loadings below 5.0 wt% yielded more accurate dosing than higher mass loadings. At 2.5 wt% and 1.5 wt% MOF in <106 μm beads, the mass loading was 86% and 103% of the expected mass fraction, respectively. The precision was slightly better than 5.0 wt%, with coefficients of variation of approximately 4.7% and 5.9%, respectively. Though these lower mass loadings exhibited more favorable reactor dosing accuracy and precision, the additional glass beads per unit mass of MOF may be a drawback for select applications and should be considered carefully in the context of the chosen application.

Characterization of MOF Structure following Dispersion

Plate-reader analysis enables rapid characterization of hundreds of reactor dosing trials but is inherently destructive and reports only on linker concentration after digestion. It cannot confirm whether the MOF framework survives the dispersion process intact. As an initial probe of how the dispersion process affects the MOF, scanning electron microscopy (SEM) was employed (Figure 3b-e). The SEM micrographs reveal that the MOFs do not strongly adhere to the glass bead surface but are instead distributed loosely among the beads (Figure 3b and c) The micrographs showed that MOF crystallites are mostly the same as before dispersion, showing only minor cracking in a fraction of the crystals. This contrasts with ChemBead formulations of molecular adsorbates, which typically exhibit more pronounced surface adhesion following preparation.³³ The weak interaction is consistent with prior studies indicating that binders are generally required for stable MOF attachment under dynamic adsorption conditions.^42-44 It may also explain why smaller beads are necessary for satisfactory reactor dosing performance with MOFs. The poorly adsorbed MOF may settle to the bottom of the dispersion if it is poorly adhered. Or, the smaller beads have a higher surface area that better promotes adhesion relative to the vessel surface.

Figure 3.

(a) general scheme showing dispersion process for Ni-MOF-74 with a picture of the final dispersion (b) SEM micrograph of Ni-MOF-74 dispersed in <106 μm beads (c) SEM micrograph showing the poor adhesion of MOF to the surface of the MOFBead (d) Ni-MOF-74 before dispersion (e) SEM micrograph of Ni-MOF-74 after dispersion (f) PXRD of Ni-MOF-74 before and after the dispersion process and (g) N₂ adsorption isotherms at 77 K of Ni-MOF-74 before dispersion in beads and on beads. The gas uptake and surface area is normalized to the expected mass of MOF based on theoretical weight fraction as the glass beads had a negligible surface area (<5m² g⁻¹).

Analysis of the dispersed MOFs by PXRD and N₂ adsorption isotherms showed that the material’s bulk crystallinity and porosity were largely retained. As shown in Figure 3f, the PXRD of MOF purified from the beads confirmed that the MOF was still crystalline. The nitrogen adsorption isotherms of the MOFs on beads showed only a modest reduction in porosity (from 1262 m² g⁻¹ to 1033 m² g⁻¹). However, this measurement assumed perfect 5.0 wt% loading on the beads. The true surface area of the MOF is likely higher as our plate reader results suggest that at least some of the MOF is not adequately dispersed in the beads. Collectively, these results indicate that although dispersion induces some minor local morphological damage and potentially modest reductions in apparent surface area, the overall structure and accessible porosity of Ni-MOF-74 are mostly preserved under the conditions tested.

Generality of Bead Dispersion Approaches

To evaluate the generality of this approach, six additional MOFs (NU-1000, HKUST-1, UiO-67-bpy, NiBDP, MOF-808, and MFU-4l), were subjected to the same dispersion and dosing procedure. These frameworks were selected as representative examples spanning several major MOF classes- carboxylate, salicylate, triazolate, and pyrazolate linkers. As shown in Figure 4b, 5.0 wt% dispersions in <106 μm glass beads yielded consistent reactor dosing across all materials, with standard deviations below 15% which is consistent with other bead-based approaches.^32,33

Figure 4.

(a) Stylized depictions of the MOFs evaluated in the MOFBead scope and their surfaces areas before and after dispersion. (b) Boxplots showing the reactor dosing accuracy and variability (n= 20 or 24 trials) of the MOFBead scope using 5.0 wt% MOF dispersed in <106 μm glass beads measured out with a 20mg scoop. The MOF-808 reactor dosing (n=8 trials) was measured using ¹H NMR against an internal standard because the UV absorbance was outside of the plate reader’s spectral range.

The porosity of the MOFs before and after dispersion was also examined. BET surface areas were maintained within 20% of their pristine values with the exception of MOF-808 (Figure 4a). In contrast, MOF-808 exhibited a substantial decrease in surface area (1586 to 1182 m² g⁻¹), indicating partial framework degradation under these conditions. Again, these apparent surface areas were collected with the MOF on beads and assume a 5.0 wt% loading. Their actual surface areas are likely higher due to material losses in the dispersion process.

Crystallinity retention was further confirmed by PXRD, which showed that all MOFs preserved their long-range order after dispersion. Together with surface area analysis, these results indicate that the dispersion method maintains the essential structural features of most MOFs tested, while highlighting framework-dependent differences in stability, in the case of MOF-808.

Application in High-throughput Screening

As a proof-of-concept study, we next sought to apply this platform to identify a novel MOF C─H oxidation catalyst in the M–MFU-4l–X series, where M = divalent first row transition metal and X is an anion (e.g. Cl⁻, or OAc⁻). The MFU-4l MOF platform was selected for this study because its Zn₅(triazolate)₆ Kuratowski-type secondary building unit (SBU) possesses four Zn centers accessible to the internal micropores and is readily amenable to post-synthetic modification via cation exchange at peripheral metal sites and anion exchange of their bound ligands.^{37, 40} Prior studies have shown that these sites participate in diverse catalytic transformations, including alkene dimerization, polymerizations, biomimetic C─H oxidation, and that catalytic performance depends the metal–anion pairing.^40,45-49 However, identifying active pairings by conventional workflows is cumbersome, emphasizing the need for a high-throughput screening platform such as the one developed here.

All permutations of M-MFU-4l-X (M = Mn, Co, Ni, Cu; X = Cl⁻, Br⁻, OAc⁻, BF₄⁻, NO₃⁻ or OTf⁻) were synthesized by cation exchange of Zn-MFU-4l-Cl followed by anion exchange (Figure 5). This process is, however, labor intensive, requiring multiple washes of the material after each exchange. To minimize the work of preparing 24 unique materials and reduce significant material demands, a semi-automated, small-scale protocol was developed (see page S27 for details). This approach enabled the preparation of all 24 unique cation and anion combinations materials in parallel. After the ion exchanges, the MOFs were dispersed on glass beads (<106 μm glass beads) and were characterized by XRF to determine final metal content.⁵⁰

Figure 5.

Overview of synthetic approach for M-MFU-4l synthesis via sequential cation and anion exchange at the SBU.

The benzylic C─H oxidation activity of the 24 separate MOFs was next assessed using 1-bromo-4-n-butylbenzene as a model substrate in MeCN with four common terminal oxidants: O₂ (air), H₂O₂, tert-butyl hydroperoxide (TBHP), and PhI(OAc)₂ (PIDA). This activity screen used approximately 25 mg of MOF for 96 reactions. Product formation was assessed by quantitative ¹H NMR against an internal standard. As shown in Figure 6, only PIDA and tBuOOH afforded significant activity. The Mn-MFU-4l derivatives in conjunction with PIDA showed the highest activity with Mn-MFU-4l-Cl performing the best, turning over 16 times per Mn center. The Co- and Ni-substituted analogues showed measurable activity with these oxidants but were uniformly worse than Mn derivatives and were not pursued further.

Figure 6.

Initial activity screen of four oxidants and 24 M-MFU-4l-X derivatives in MeCN at 65 °C. Turnover numbers were normalized against non-Zn metal content as determined by x-ray fluorescence.

Having identified Mn-MFU-4l-X derivatives as the most promising catalysts, we next expanded the catalyst screen to determine activity in additional solvents, as well as with additional oxidants known to effect C─H oxidation with Mn catalysts. The outcome of this screen is shown in Figure 7. Again, the hypervalent iodine reagent PIDA in MeCN gave the most ketone product (TON=7).⁵¹ Other solvents oxidants and solvents were generally less effective, yielding little to no conversion. Only TBHP gave reasonable activity. Importantly, control reactions clearly indicate that the observed oxidation activity is catalyzed by the Mn centers in the MOF. Analogous reactions using Zn-MFU-4l-Cl, MnCl₂, or just the glass beads showed little to no activity, indicating that the reaction is likely catalyzed by SBU-based Mn centers.

Figure 7.

(a) Heatmap showing relative activity of Mn-MFU-4l-Cl using eight terminal oxidants and five solvents. Background oxidation activity of the oxidants in MeCN without MOF catalyst is shown in the sixth column. 3.5 equivalents of oxidant were used in all reactions with the exception of air. (b) Additive screen with added H2O and AcOH. Equivalents for each is relative to 1-bromo-4-n-butylbenzene starting material.

The catalytic oxidation reaction using PIDA was selective for the ketone-containing product. No evidence for acetoxylated or alcohol products, which may be initially expected when using PIDA as a oxidant, were noted under the reaction conditions evaluated in Figure 7a. As such, we hypothesized that the active oxidant was in fact iodosyl benzene formed through a minor equilibrium with adventitious water. This equilibrium is known to be important for other Mn-based C─H oxidation catalysts when using PIDA as the terminal oxidant.^52,53 However, if this equilibrium is pushed too far toward iodosyl benzene, an insoluble polymer forms.⁵⁴ Thus, we reasoned that equilibrium would therefore need to be carefully controlled. To test the role of water and AcOH additives on catalyst efficiency, we next conducted a parallel optimization screen of water and acetic acid additives in Mn-MFU-4l-catalyzed oxidation of 1-bromo-4-n-butyl benzene with PIDA. The results of this experiment are shown in Figure 7b. Consistent with other Mn oxidations using PIDA as the terminal oxidant, the reaction showed a mild dependence on added water with a marginal improvement in activity from 7 to 9 TON using 7 equivalents of water and no added acetic acid.⁵¹ Additional water beyond 7 equivalents or any acetic acid significantly lowered the activity of the catalyst.

In a final set of experiments, we next sought to establish the reproducibility of the MOFBead dispersions and probe if the glass beads significantly altered the reaction outcome. Repeating the reaction 12 times in parallel gave relatively small variation in activity. The TON of all trials was just +/− 1 turnover of the average. The effect of the glass beads on the total TON was also small. Manually setting up the reaction without beads (but at a larger scale to facilitate accurate measurement of catalyst), gave an identical yield to the small-scale reaction with MOF-Beads. Moreover, adding additional glass beads to the reactors did not significantly change the catalyst activity. These observations suggest that this specific transformation is not significantly affected by the presence of <106 μm glass beads.

Although benzylic C─H oxidation can be more efficiently accomplished using alternative methods, this study illustrates the potential of higher-throughput workflows in MOF catalysis. In this proof-of-concept study, approximately 150 reactions were evaluated using less than 50 mg of MOF total. Importantly, screens of 48 reactions generally took fewer than 15 minutes to dose the reaction vials. Such large-scale optimization campaigns would be difficult to accurately conduct using traditional, labor-intensive manual dosing or costly automated powder dispensers.

4. CONCLUSION

In summary, this study describes the synthesis, optimization, and implementation of MOFBead dispersions for low-cost, accessible, catalytic screening. Our results demonstrate that bead dispersion approaches can work for MOF catalysis but adapting established ChemBead methodologies to MOF catalysis is not straightforward. Unacceptably high variability in volumetric reactor dosing was observed using conventional formulations, highlighting the need for MOF-specific optimization. Through systematic refinement aided by parallel characterization with a plate reader, smaller, unfunctionalized glass beads were identified as the most effective dispersants. This approach was shown to facilitate accurate and precise reactor dosing down to 100 μg of MOF per reaction using inexpensive 3D-printed scoops.

The utility of the optimized dispersions was demonstrated through a crude activity assay evaluating a series of cation- and anion-exchanged M-MFU-4l derivatives. The C─H oxidation performance of these MOFs was surveyed using four different oxidants. Among these, Mn–MFU-4l catalysts in combination with PIDA exhibited notable activity for benzylic C─H oxidation under mild conditions. Further evaluation of solvent, oxidant, and additive stoichiometries yielded modest improvements in catalytic performance. Overall, this proof-of-concept study emphasizes the potential of this platform for more efficient catalyst optimization workflows. Notably, over 150 reactions were conducted in these screening experiments using less than 50 mg of MOF material, a figure that has potential to be reduced by using smaller scoops.

While this MOFBead reactor dosing approach offers a low-cost and accessible strategy, it is not without limitations. The approach described here is not likely universally applicable. Of the seven MOFs evaluated, at least one exhibited significant structural degradation during the dispersion process, suggesting that other frameworks may be similarly vulnerable. Nevertheless, the use of plate-reader-based characterization enables rapid evaluation of dispersion stability and dosing accuracy, potentially facilitating the development of modified preparation protocols tailored to more sensitive materials. A further limitation of the method is the potential influence of mechanical agitation from the glass beads during catalysis. In this study, comparable catalytic yields were obtained with and without beads, although this outcome is unlikely to be fully generalizable across all MOFs and reaction types. Despite these constraints, the MOFBead strategy substantially reduces material consumption and accelerates catalyst screening, representing a significant advance for the evaluation of MOF-based catalysts, particularly those derived from complex or resource-intensive syntheses. Applications of this approach to the identification of novel MOF catalyzed transformations is underway in our laboratory.

Supplementary Material

The Supporting Information is available free of charge on the ACS Publications website.

Additional experimental details, characterization, and methods, including photographs of dispersion preparations (PDF)

ABBREVIATIONS

DMF

dimethylformamide

MeOH

methanol

DMSO

dimethyl sulfoxide

EtOH

ethanol

Et₂O

diethyl ether

DCM

dichloromethane

MeCN

acetonitrile

THF

tetrahydrofuran

DMA

dimethylacetamide

EtOAc

ethyl acetate

TFE

2,2,2-trifluoroethanol

HFIP

1,1,1,3,3,3-Hexafluoropropan-2-ol

IDTS

2-(tert-butyl-sulfonyl)iodosylbenzene

TBHP

tert-butyl hydroperoxide

Di-TBP

di-tertbutyl peroxide

PIDA

iodobenzene diacetate

PIFA

Bis(trifluoroacetoxy)iodobenzene

H₂O₂·urea

urea hydrogen peroxide adduct

H₂O₂

hydrogen peroxide, 30 wt% in water