The active sites in catalysts are often sensitive to their surrounding environments. Consequently, active site engineering has been developed to allow the modulation of a catalyst environment to exploit or inhibit activity (Pikus et al., 1997; Garce et al., 2009; Nasalevich et al., 2014; Liu et al., 2018; Reiner et al., 2019); however, tuning of the solid support, as in single-site heterogeneous catalyst systems, is often difficult to achieve. For this reason, metal-organic frameworks (MOFs) with a high degree of tunability are attractive support materials (Vermoortele et al., 2012; Vandichel et al., 2015; Liu et al., 2018; Palmer et al., 2018). Constructed from metal or metal cluster nodes linked via multitopic organic ligands, MOFs are an increasing popular class of designer materials due to the ability to obtain highly crystalline, porous, and stable structures (Yaghi et al., 2003; Kitagawa et al., 2004; Furukawa et al., 2013; Jiao et al., 2018; Chen et al., 2019; Yang and Gates, 2019).

One ubiquitous class of materials which has benefitted from heterogenization and immobilization on solid supports is polyoxometalates (POMs) (Long et al., 2007; Frenzel et al., 2016; Hou et al., 2018). POMs are isolated metal oxide clusters, whose tunable size, charge density, and acid strength allow for a wide range of applications, primarily in catalysis (Pope, 1983; Wang and Yang, 2015; Anyushin et al., 2018; Gumerova and Rompel, 2018; Sullivan et al., 2018). However, POMs suffer from low surface areas, high solubility in polar solvents, and limited stability in aqueous environments, making POMs difficult to exploit as reusable catalysts (Kholdeeva et al., 2010). To increase surface area and impart some stability, POMs can be supported on porous heterogeneous materials, among which are metal-organic frameworks (MOFs) (Salomon et al., 2018; Wang et al., 2018; Zhong et al., 2018).

As support materials (Sosa et al., 2018), MOFs can interact with POMs and other guest molecules via chemical bonds, steric encapsulation within pores, or electrostatic interactions. With chemical bonding as found in polyoxometalate open frameworks, POMs serve as the nodes of the framework (Nohra et al., 2011; Miras et al., 2014), and consequently the active sites are blocked by the structural linkers. Steric trapping relies on a MOF whose pores can accommodate a POM, but whose apertures are much smaller than the POM thus preventing POM diffusion out of the MOF (Song et al., 2011; Salomon et al., 2015; Cai et al., 2019). This aperture dependence was investigated in a recent report (Lin et al., 2018). Lastly, functionalities in MOFs can impart favorable electrostatic interactions (Férey et al., 2005). For example, in Zr-based MOFs, the nodes have an affinity for oxyanions (Howarth et al., 2015). Without these weak interactions, the POM guests are able to freely diffuse in and out of the MOF, making a composite structure difficult to isolate and recycle.

NU-1000 is a Zr-based csq-net MOF comprised of mesoporous 31 Å hexagonal channels and microporous 12 Å triangular channels (Figure 1). We have recently reported the impregnation of NU-1000 with phosphotungstic acid (H₃PW₁₂O₄₀) resulting in the composite material, PW₁₂@NU-1000, which was stable to leaching (Buru et al., 2017, 2018; Ahn et al., 2018) due to favorable electrostatic interactions. When the material activated (removing the solvent guest molecules) using supercritical CO₂ (scCO₂) drying, the [PW₁₂O₄₀]³⁻ clusters were found in the easily accessible mesopore. Evidence of this location was observed in the pore-size distribution calculated using the N₂ isotherm and confirmed via difference envelope density analysis, where the high intensity, low angle peaks of a powder X-ray diffraction (PXRD) pattern generated from a synchrotron X-ray source were used to create an electron envelope from with the known parent MOF was subtracted and the residual electron density mapped onto the parent MOF indicated the location of the guest molecules (Yakovenko et al., 2014; Gallington et al., 2016). Upon heating the scCO₂-dried material, the [PW₁₂O₄₀]³⁻ clusters migrate from the mesoporous channel to the microporous channel, likely owing to the enhanced electrostatic and van der Waals interactions when cited in the micropore. The movement was also accompanied by the loss of water weight from thermogravimetric analysis (TGA). We proposed that this movement occurs because the POM can fit in the c-pores, the 8 × 10 Å windows between the channels, since this transformation readily occurs in the absence of solvent. Different activations of these materials yielded different rates and selectivities in the oxidation of 2-chloroethyl ethyl sulfide (CEES), a mustard gas (bis(2-chloroethyl)sulfide) simulant (Figure 2). Interestingly, due to mechanistic differences (Al-Ajlouni et al., 2007; Lousada et al., 2012), the homogeneous H₃PW₁₂O₄₀ promoted the selective oxidation of CEES, while the MOF alone catalyzed the over-oxidation. This allowed for insight into which clusters were more accessible in the composite material through analysis of reaction selectivity.

To determine a more exact nature of this migration and the effect the POM location has on catalysis, we employ here two MOFs, NU-1000-NDC and NU-1008 (Figure 1) (Lyu et al., 2018; Peters et al., 2018) to immobilize [PW₁₂O₄₀]³⁻ clusters. These MOFs have the hierarchical channel-type structure that NU-1000 possesses, except the apertures connecting the channels are blocked in NU-1000-NDC (c-pores in Figure 1) or are smaller in NU-1008 (c- and c′-pores in Figure 1). This would inhibit [PW₁₂O₄₀]³⁻ movement between channels if the mechanism of POM migration was allowed by movement through these windows, so that the composites activated via supercritical CO₂ drying and via heating under vacuum should only differ in the amount of physisorbed water. The influence of this residual water is investigated in the catalytic oxidation of CEES, using hydrogen peroxide.

NU-1000-NDC and NU-1008 were selected for this study to investigate whether the [PW₁₂O₄₀]³⁻ moves through the c- and c′-pores (Figure 1) as previously proposed. In NU-1000-NDC, 2,6-naphthalene dicarboxylate (NDC) bridges two nodes in the c-pore, effectively reducing the size of that aperture (Peters et al., 2018). Both the c- and c′-pore diameters in NU-1008 are reduced in size compared to those in NU-1000-NDC (Lyu et al., 2018; Wasson et al., 2018). If the mechanism involved complete desorption from the mesopore and re-adsorption into the microporous channel of the POM, the size and accessibility to the c/c′-pores would have negligible effect on POM movement.

PW₁₂@NU-1000-NDC and PW₁₂@NU-1008 were prepared in a similar manner to PW₁₂@NU-1000 (Buru et al., 2017). Briefly, the MOFs were soaked in an aqueous H₃PW₁₂O₄₀ solution for 3 days, followed by rigorous washing with water and ethanol. After soaking in ethanol overnight, the composites were subjected to supercritical CO₂ drying. Half the material was then heated to 120°C for at least 2 h and designated PW₁₂@NU-1008-120°C and PW₁₂@NU-1000-NDC-120°C, while the other half was used “as-is” for subsequent characterization and designated PW₁₂@NU-1008-scCO₂ and PW₁₂@NU-1000-NDC-scCO₂.

Inductively coupled plasma optical emission spectroscopy (ICP-OES) analyses of the digested MOFs show that [PW₁₂O₄₀]³⁻ does not leach during heating. The loadings were found to be 0.7 PW₁₂/Zr₆ node for NU-1000-NDC and 0.8 PW₁₂/Zr₆ node in NU-1008. Scanning electron microscopy (SEM) images reveal that the size and shape of the MOF crystallites remains the same before and after POM loading (Figures S1, S2). Energy dispersive X-ray spectroscopy (EDS) line scans display that the POM distribution in the MOF is homogeneous throughout the crystallites.

From the TGA traces (Figure S3) of the materials under nitrogen, both scCO₂-activated MOFs lost ~5 wt% more than the 120°C-activated counterparts at 250°C. This initial loss has been attributed to residual water physisorbed to the POM species, which remain after removal of solvent. Once heated to 120°C, the elevated temperature activation step removes the physisorbed water which was difficult to remove from the POM after scCO₂-drying alone. These residual waters were factored into catalyst loading calculations.

In the N₂ sorption isotherms (Figures S4, S5), the maximum uptakes of the composites are significantly lower gravimetrically than the parent MOFs. Since the POMs constitute >50 wt% according to ICP analyses, volumetric equivalents were obtained after estimating densities to determine whether the decreased uptake was a result of mass gain or the degradation of the MOF (Figure 3). For NU-1000-NDC, the MOF has a Brunauer-Emmett-Teller (BET) area of 1,150 m²/cm³ which slightly decreases to 990 m²/cm³ in PW₁₂@NU-1000-NDC-scCO₂ and 1,100 m²/cm³ for PW₁₂@NU-1000-NDC-120°C. Similarly, in NU-1008 the MOF has a BET area of 910 m²/cm³, and the scCO₂ and 120°C activated composites have areas of 820 and 880 m²/cm³, respectively. With these marginal decreases in BET areas, the decreased gravimetric uptake can be attributed primarily to the additional mass of the POMs and not the degradation of the MOF. In the density functional theory (DFT)-calculated pore size distribution (PSD) for NU-1000-NDC, the mesopore around 30 Å decreases after POM loading and scCO₂-drying (Figure 3C). Once heated to 120°C, some of the mesopore volume is regained. This hints that the POM occupies the mesopore after scCO₂ drying and some POM leaves after heating. The PSD for NU-1008 shows a similar behavior from the MOF to the scCO₂-dried sample (Figure 3D); however, after heating the mesopore volume is retained, suggesting that the POMs are unable to move from the mesopore.

The PXRD patterns for the parent NU-1000-NDC and NU-1008 are similar (Figure S6); the most prominent peaks arise from the (100) plane around 2θ = 2.5° and (200) peak around 5°. Once the POM is installed, the highest intensity is the (2–10) Bragg peak appearing around 2θ = 4°, indicating the extra electron density from the POMs resides in the mesopore (Buru et al., 2018). ScCO₂ drying has minimal effect on the composites PXRD pattern (Figure 4), suggesting little to no POM movement during activation. After heating to 120°C for 2 h, neither composite appears to have changed. However, after 12 h of heating at 120°C, the PW₁₂@NU-1000-NDC closely resembles the parent MOF, indicating that the POMs now occupy the micropores (Figure 5). This change is indicative of the c′-pores being flexible enough to allow POM movement. In PW₁₂@NU-1000, complete movement of the POM from the meso- to micropores occurs, suggesting that PW₁₂@NU-1000-NDC requires more energy to undergo the same transformation, so while blocking the c-pore does not prevent the POM from moving, it does force the POM through a higher energy barrier intermediate. Interestingly, PW₁₂@NU-1008 appears to have retained its PXRD pattern under the same heating conditions, indicating the POMs remain in the mesopore (Figure 5) and supporting the previously proposed mechanism where the POMs move through the c- or c′-pores. In situ variable temperature PXRD patterns (Figure S7) show this transformation occurs gradually over time.

The oxidation of CEES was selected as a model reaction (Figure 2). The catalytic setup consists of 1 mol% catalyst (0.86 μmol active clusters, i.e., a node or POM) in a vial suspended in acetonitrile and heated to 45°C. CEES and internal standard were then added once the vial reached the set temperature. Then to start the reaction, 1.5 eq of 30% aqueous H₂O₂ was added. The disappearance of CEES was monitored by GC-FID. From the kinetic traces (Figure 6), the H₃PW₁₂O₄₀, NU-1008, and NU-1000-NDC alone exhibit similar rates, achieving ~50% conversion after 30 min. With the same number of active clusters, PW₁₂@NU-1000-NDC and PW₁₂@NU-1008 convert CEES at faster rates than the individual components alone. For PW₁₂@NU-1000-NDC (Figure 6 left), the reactivity trend is consistent with PW₁₂@NU-1000, where the scCO₂-activated sample goes to full conversion at the highest rate, and the 120°C-activated sample has slightly lower activity. The marginal difference in initial turnover frequencies (TOF; 25 and 29 min⁻¹ for scCO₂- and 120°C-activated, respectively) suggests that the POMs exist in both meso- and microporous channels; this can be compared to the 3-fold difference in initial rates in NU-1000 when the POM is located exclusively in one or the other channel. Interestingly, PW₁₂@NU-1008-scCO₂ is slower for CEES oxidation than PW₁₂@NU-1008-120°C, exhibiting TOFs of 14 and 21 min⁻¹, respectively. In this case, the POMs are located only in the mesopore, unable to migrate to the micropores. This finding suggests that the removal of water when not accompanied by POM movement leads to rate enhancement, and that limiting POM migration to a more thermodynamically favorable site allows for better accessibility of the active sites. Note, the apparent enhanced rate of the NU-1000-NDC over NU-1008 can be attributed to particle size differences and the poor catalyst accessibility of the substrate from the micropore, both hindering substrate diffusion.

Two MOFs, NU-1000-NDC and NU-1008 were used to immobilize H₃PW₁₂O₄₀. The composite materials were activated via supercritical drying and/or heating to 120°C, resulting in PW₁₂@NU-1008-scCO₂, PW₁₂@NU-1008-120°C, PW₁₂@NU-1000-NDC-scCO₂ and PW₁₂@NU-1000-NDC-120°C. POM location in each was confirmed via PXRD pattern analysis. For NU-1000-NDC, the POM required longer heating for POM movement to occur than in previously reported NU-1000. In NU-1008, POM movement is completely inhibited and the POM remains in the mesopore after heating. The composite materials were more active than the individual components alone. Among the composites, the location of POM and water content in composite was found to affect the oxidation rate of a mustard gas simulant, CEES. In NU-1000-NDC, the [PW₁₂O₄₀]³⁻ cluster in the mesopore is more active than when the material is heated and the cluster resides in the micropore. In contrast, when comparing the NU-1008 composites where both materials have the [PW₁₂O₄₀]³⁻ cluster in the mesopore, heating the sample to remove water yields the more reactive catalyst. These findings highlight the importance of active site location on a heterogeneous support, and future studies aim to exploit the effect of POM location on catalysis.

All datasets generated for this study are included in the manuscript and/or the Supplementary Files.

All experimental work was performed by CB, JLy, and JLi under guidance of OF. Data analysis was performed by CB with valuable contributions from OF. The manuscript was written by CB with contributions and corrections from JLy, JLi, and OF.