**2. Results**

MIL-101(Fe) was obtained by a solvothermal method from benzene-1,4-dicarboxylic acid and iron (III) chloride, according to the method of Zhu et al. [43] The powder X-ray diffraction (PXRD) pattern of the material prepared at 3 hours reaction time confirms the identity of the material by comparison with the simulated pattern from the literature (Figure 3). The differences in relative intensity between the low angle peaks (at less than 6 ◦2θ ) and those at higher angle (greater than 8 ◦2θ ) in the measured pattern compared to the simulated pattern is accounted for by the presence of solvent in the as-prepared sample; in contrast, the simulated pattern is for an idealised guest-free framework. These X-ray powder patterns are typical of MIL-101 samples reported in the literature [44–46]. There is also evidence in the PXRD for the presence of an impurity phase in the sample, which we identify below as being MIL-88B (Figure 1) [47]. It is already known that MIL-88B and MIL-101 are "framework isomers" and may form in competition to each other [4,48,49].

**Figure 3.** Powder X-ray diffraction patterns of samples of MIL-101 prepared by solvothermal synthesis at 3 hour reaction times. The simulated pattern of MIL-101 was produced from the published structure of <sup>F</sup>érey et al. [36] and that of MIL-88B from the refined diffraction profile in this work for MIL-88B(Fe).

Figure 3 also shows PXRD patterns of materials prepared where small amounts of Sc3+ are used to replace some of the Fe3+, or some of the BDC linker is replaced by mono-2-sulfobenzene-1,4-dicarboxylate (MSBDC). In both cases, the crystallinity of the resultant product is considerably less than for the Fe material, and the MIL-88B impurity remains in the samples. SEM was used to detect the presence of the sulfo-modified ligand in the material, but no signal from sulfur was observed. Since we used this method successfully to image sulfur in sulfo-modified UiO-66 [31], we conclude that for MIL-101(Fe) prepared by this method, the incorporation of the sulfo-modified ligand is negligible. Fourier transform infrared (FT-IR) spectra also provided no evidence for any characteristic sulfo bands. Owing to the lack of any sulfo-functionalisation, we excluded these samples from any further study.

The highest scandium amount that can be introduced to MIL-101(Fe) is only 10%. If 50% of the Fe is replaced by Sc (note that this is the nominal metals ratio used in the synthesis) then the material formed is solely MIL-88B under the synthesis conditions that we have used, labelled MIL-88B(Fe0.5Sc0.5), and indeed if longer synthesis times are used then, even for the iron material, only MIL-88B is found (see Figure 4). The powder X-ray pattern of MIL-88B matches that reported by Ma et al. who prepared the iron form by microwave synthesis in the same solvent [50]. Figure 4 also demonstrates the stability of MIL-88B in DMSO upon prolonged heating: although the powder pattern shows some broadening of the peaks, the material clearly has the same structure. The materials are, however, unstable in water: upon heating to 140 ◦C for 3 hours, the MOF structure collapses and the only crystalline

product recovered is the acid of the BDC linker, along with an orange powder, presumably amorphous iron oxide.

**Figure 4.** Powder X-ray diffraction patterns of MOFs transformation from MIL-101(Fe), 3 hour crystallisation) to MIL-88B(Fe), 40 hour crystallisation, along with mixed (Fe,Sc) (nominal metals ratio 1:1), 3 hour crystallisation. The second from top pattern illustrates the stability of the mixed-metal phase on heating in DMSO.

Confirmation of the identity of the more stable MOF structure in the synthetic conditions used was provided by a more detailed analysis of higher resolution powder X-ray diffraction patterns, and Figure 5 shows the profile fits to the MIL-88B(Fe), MIL-88B(Sc0.5Fe0.5) and MIL-88B(Sc) materials. The refined lattice parameters (Figure 5) fitted using the hexagonal space-group *P*62*c* are consistent with those reported in the literature for MIL-88B materials: for example, Ma et al. found *a* ~ 12.7 Å, *c* ~ 18.4 Å and *V* ~ 2570.2 Å3 for MIL-88B(Fe) isolated from *<sup>N</sup>*,*<sup>N</sup>*-dimethylformamide (DMF) [50], and Mowat et al. found *a* = 11.2190 Å, *c* = 19.373 Å, *V* = 2112.27 Å3 for MIL-88B(Sc) containing a mixture of water and DMF [39]. MIL-88B, however, is an example of a breathing MOF and the lattice parameters can vary dramatically with the amount, or type, of solvent present [51], so it is difficult to draw conclusions in the trends in the values of the fitted lattice parameters for the three materials that we have studied. Larger unit cell volumes are observed for both the Sc-containing samples we have prepared compared to the iron form, which is consistent with the larger ionic radius of octahedral Sc3+ compared to Fe3<sup>+</sup> [52]. However, the presence of different amounts of solvents (or mixtures of water and DMF) cannot be ruled out, which may also adjust the unit cell volume.

**Figure 5.** Profile fits to powder X-ray diffraction patterns of (**a**) MIL-88B(Sc), (**b**) MIL-88B(Fe0.5Sc0.5), and (**c**) MIL-88B(Fe) (Cu *K*α1 radiation) fitted using space-group *P*62*<sup>c</sup>*. The points are the measured data, the red line is the final fit, the blue line is the difference curve and the pink ticks are the positions of the allowed reflections. Refined lattice parameters and unit cell volumes are provided on each panel.

In addition to the different unit cell of the mixed-metal material, more substantive evidence for the formation of a mixed-metal sample of the MOF, rather than a phase-separated mixture of materials, comes from EDX mapping on the SEM, Figure 6. This shows how the two metals are both present in micro-sized crystallites of MIL-88B(Sc0.5Fe0.5). Proving the degree of mixing of the two metals on smaller length-scales is challenging; in the previous work on MIL-100(Fe,Sc), which contains the same trimeric building units as found in MIL-88B, it was impossible to determine, using extended X-ray absorption fine structure (EXAFS) spectroscopy, whether the individual clusters contained mixtures of metals or whether single metal clusters were present [41].

**Figure 6.** Elemental maps of MIL-88B(Fe0.5Sc0.5) at two magnifications showing the distribution of the two metals, and the presence of chlorine.

The elemental mapping suggests that the concentration of iron in the MIL-88B(Fe,Sc) sample is lower than that of scandium, despite the initial 1:1 molar ratio used in synthesis. This was verified by bulk analysis for metals using ICP-OES, which gave a Sc:Fe molar ratio of 3.1:1. X-ray photoelectron spectroscopy (XPS, Figure 7), was used to quantify the metal ratio at the surface of the material, and integration of the signal gave a Sc:Fe ratio of 6.0:1. The typical depth of the XPS analysis for these elements is around 5–10 nm, which shows how the surface of the particles is scandium-rich. First, this suggests that most of the Fe is present within the crystallites of the MIL-88B, and, second, shows how the surface is terminated substantially by Sc-containing species, which are strongly bound, since they are not removed by the washing used in the sample preparation. One possible explanation for the excess scandium at the surface is the presence of a scandium chloride species, necessary for charge-balance of the terminated crystal surfaces. Some of the XPS signal can indeed be accounted for as ScCl3 (Figure 7a), and this result is also consistent with the presence of chlorine detected in the EDXA mapping on the SEM. The majority of signal (>80%) in the Sc <sup>2</sup>*p*3/2 region (Figure 7a), however, can be accounted for by the presence of octahedrally coordinated scandium, assigned using ScOOH as a reference, as expected for the MIL-88B structure. The Fe <sup>2</sup>*p*3/2 region of the XPS (Figure 7b) is more complex due to multiplet splitting [53–56]. Careful analysis of the overall Fe <sup>2</sup>*p*3/2 envelope necessitated a component at around 709 eV, indicating the presence of a significant amount of Fe2<sup>+</sup> as well as Fe3+. This is unlike the spectrum of the single metal MIL-88B(Fe) (Supplementary Materials) that shows >90% Fe3+, as expected, and would sugges<sup>t</sup> that the mixed-metal material contains defects (see below for a further discussion of this).

**Figure 7.** X-ray photoelectron spectra of MIL-88B(Fe,Sc) in (**a**) the Sc 2*p* region and (**b**) the Fe <sup>2</sup>*p*3/2 region. The fitted contributions are shown with reference spectra in the literature used for guidance (see text).

The FT-IR spectra of the MIL-88B materials (Figure 8) shows all the vibrational bands characteristic of framework-coordinated carboxylate groups expected in the MIL-88B structure, and further evidence for the isostructural nature of the samples (see Supplementary Materials for assignment of the spectra). Some weak additional bands in the spectrum of the MIL-88(Fe,Sc) sample sugges<sup>t</sup> the presence of some uncoordinated (protonated) carboxylic acid groups. This is not due to the presence of crystalline H2BDC (not seen in the powder XRD, Figure 5, and which also has a different FT-IR spectrum), and therefore suggests a defective structure for the mixed-metal material, with some linkers that do not bridge pairs of clusters. The thermogravimetric analysis (TGA) data (Supplementary Materials) show that all the solvents and water molecules in MIL-88B materials are removed at above ~150 ◦C. On continued heating, a decomposition of the MIL-88B network occurs above 450 ◦C, forming Fe2O3 and/or Sc2O3 as the final product, which is consistent with the literature [47]. The TGA of the mixed-metal material (Supplementary Materials) shows an extra inflection in mass loss in the region corresponding to the combustion of the linker, which is consistent with the presence of some partially coordinated benzene-1,4-dicarboxylate groups. This is consistent with the more defective structure of this material, also indicated by IR, and charge balance in the material may be explained by the presence of some Fe2<sup>+</sup> that was seen by XPS, as discussed above.

**Figure 8.** Infrared spectra of MIL-88B materials. The features labelled \* are assigned as due to uncoordinated carboxylic acid groups.

Three samples were selected to examine their effectiveness as catalysts for glucose conversion: MIL-88B(Fe), MIL-88B(Fe,Sc), and MIL-88B(Sc). Given the instability of the materials in water, DMSO was used as the reaction solvent, in which the materials proved to be stable (Figure 4). The catalytic results are summarised in Table 1, which shows the total glucose conversion as well as the three desired products analysed for: fructose, mannose, and 5-HMF. We report results from two different catalyst loadings (based on mass, 30:1 and 7.5:1, glucose to catalyst ratio) and at two different temperatures (120 ◦C and 140 ◦C). Firstly, it is important to note that DMSO alone, in the absence of any catalyst, results in a conversion of just 1.6% at 120 ◦C and 60.3% at 140 ◦C. However, the product mixture contains only a minor amount of 5-HMF at the upper reaction temperature and neither of the intermediate sugars, fructose or mannose, are present, which suggests a decomposition pathway that does not yield these desired products. Indeed, Jia et al. found that anhydrous DMSO alone gave cellobiose as the main dehydration product of glucose at 130 ◦C for 2 hours [57]. In the presence of the MIL-88B solids, a much more significant formation of 5-HMF is seen, along with the intermediate fructose and mannose sugars. DMSO is known to act as a catalyst for the dehydration of fructose to 5-HMF in the absence of any Brønsted acid at 150 ◦C [58]. This implies that the MIL-88B acts as a Lewis acid catalyst that produce the fructose by glucose isomerisation, which can then be converted to 5-HMF by the DMSO solvent.

The mixed-metal MIL-88B(Fe,Sc) gives the highest conversion (70.7%) and 5-HMF yield (24.9%) at 140 ◦C with a 7.5 substrate–catalyst ratio. This catalyst thus produces significant amounts of 5-HMF from glucose. The mixed-metal material is more effective than either the MIL-88B(Fe) or MIL-88B(Sc), which may be due to greater structural disorder leading to a greater concentration of defects, such as coordinatively unsaturated metal sites, implying more Lewis acid sites. As noted above, the degree of mixing of the metals is difficult to prove unambiguously, but since the mixed-metal material always gives a higher glucose conversion than either of the single metal materials, there must be a cooperative effect, which in turn suggests that the two metals are in close proximity. In previous work on MIL-100(Sc), it was noted that upon addition of excess Fe, the presence of small amounts of amorphous Fe2O3 enhanced the Lewis acidity [41], so it is possible that we are seeing the same

beneficial effect here, although the presence of small amounts of an amorphous second phase is difficult to determine.

The effect of temperature is also significant on the catalysis. At the temperature of 120 ◦C and the catalyst-to-substrate ratio of 7.5, the selectivity towards 5-HMF is notably higher over the mixed-metal material compared to Fe-only material (16.8% vs 1.7%). At the lower catalyst-to-substrate ratio of 30:1, the mixed-metal material shows a slightly lower activity than the Fe-only material, but a notably higher 5-HMF selectivity. The varying response of the two materials to temperature at both catalyst loadings is consistent with a different, and more favourable, mode of action in the mixed-metal material. Interestingly, the scandium material has a higher selectivity to 5-HMF than the iron at the catalyst low loadings, ye<sup>t</sup> lower total conversion, providing evidence that each metal has a distinct role in the catalysis. Vlachos and co-workers studied the kinetics and mechanism of tandem Lewis–Brønsted acid catalysis of the same reaction, and reported that relative Lewis–Brønsted acid concentration has an optimum and pronounced effect on 5-HMF yield, which leads to 5-HMF yield being maximised due to a change in the rate-limiting step from fructose dehydration to glucose isomerisation [59]. In this sense, the MIL-88B(Fe) catalyst shows the formation of fructose both at 120 ◦C and 140 ◦C; however, the MIL-88B(Sc) catalyst yields higher 5-HMF yields. On the other hand, the mixed-metal MIL-88B(Fe,Sc) yields the highest 5-HMF yields, which is more clearly pronounced at 140 ◦C. Hence, the origin of the behaviour of our catalysts may be due to the distribution and availability of acid sites.

We considered the recyclability of the MIL-88B(Fe,Sc) catalyst by isolating the solid after a run, washing with DMSO, and adding fresh reagents (Figure 9). This was done at a larger scale than the screening reactions reported in Table 1 so mass transport may be different; however, the material maintains activity after three cycles and only after the fourth use shows a notable drop in glucose conversion. The recycling of MOF catalysts has not always been reported in the literature, but for the results that are available, we note that, although the recyclability of MIL-88B(Fe,Sc) is not as complete as for some sulfonated MIL-101(Cr) materials (albeit under different reaction conditions) [27,29], others have noted that the build-up of humins can make recycling difficult [10]. Nevertheless, the observation of recyclability points towards the use of the materials in continuous reactors, as has been achieved for sulfonated MIL-101(Cr) in a fixed-bed system [26]; here, conversion need not be high to generate useful quantities of the desired product.

**Figure 9.** Recyclability of MIL-88B(Fe,Sc) as a catalyst for glucose conversion (120 ◦C in DMSO for 3 hours).

In comparison to other MOF-based solid acids that have been recently investigated as catalysts for glucose conversion, the MIL-88B(Fe,Sc) shows favourable properties. Tsapatsis and co-workers have recently showed that a composite of MIL-101(Cr) and nanoparticles of Cr(OH)3 gave the highest conversion of glucose in a two-step process with ethanol as the solvent for isomerisation to ethyl fructoside, followed by the addition of water to allow hydrolysis to fructose, each performed at 100 ◦C for 24 hours: the catalyst was more effective than a physical mixture of the two component solids and

various other MIL-101(Cr) materials, but also outperformed the zeolite Sn-beta giving >75% conversion, with almost 60% of the product fructose [29]. Yabushita et al. used the zirconium MOF NU-1000 modified with phosphate to optimise the 5-HMF yield from glucose and also screened various solvent mixtures: for their catalysis, however, water-DMSO gave the lowest overall glucose conversion in comparable conditions to those we have used (at 140 ◦C for 5 hours), and THF or propanols as solvents gave almost complete conversion [30]. Looking at other solid catalysts applied in DMSO solvent, the recent work of Li et al. on MnO2 particles surface modified with Sn4+-containing coordination polymers provides a useful comparison [60]. Their catalyst achieved ~56% conversion at 150 ◦C in 5 hours, with a 60% selectivity towards 5-HMF, which was superior to other Sn-based catalysts reported in the literature in DMSO solvent. Therefore, the MIL-88B(Fe,Sc) catalyst is comparable to the Sn-based catalyst (with higher conversion, although lower selectivity), but has the advantage of being prepared in a single-step, rather than being a surface-modified composite material. Other reported systems that show a higher conversion of glucose to 5-HMF in DMSO involve a multi-step synthesis of the catalyst and strong mineral acids. This includes Sn-containing vanadium phosphate, prepared in three steps starting from a phosphoric acid treatment of vanadium oxide [61]; sulfated mesoporous niobium oxide prepared in several steps, involving ultrasound and hydrothermal treatments, with a final sulfuric acid treatment [62]; and sulfated zirconia materials that are activated using chlorosulfonic acid in ethylene dichloride [63].

The nature and location of the Lewis acid sites in MOF catalysts have not been extensively studied in glucose isomerisation reactions, and the accessibility of active sites will depend on the substrates used. Ammonia temperature-programmed desorption (TPD) was used to characterise the MIL-88B(Fe,Sc) material (Figure 10), and this confirms the presence of surface acid sites at a concentration of 0.36 mmol g<sup>−</sup>1. The results are complicated by the onset of collapse of the MOF at higher temperatures (see TGA in Supplementary Materials), but a desorption event between 100 and 300 ◦C is characteristic of the interaction of ammonia with surface acid sites. This is similar behaviour to that reported for the Cu-containing MOF-74 by Jiang et al., who observed desorption of ammonia around 100–250 ◦C under similar conditions [64].

**Figure 10.** Temperature-programmed desorption of ammonia from MIL-88B(Fe,Sc), with data from a fresh sample of the material (without ammonia adsorbed) heated under the same conditions as a blank.



While MIL-88B is a porous, breathing framework in the presence of certain solvents [51], it is likely that the bulky glucose does not penetrate into the MOF structure, but the activity is instead associated with the surface of the crystallites. Additional PXRD measurements made on samples immersed in DMSO show no evidence for pore-opening in this solvent at room temperature (Supplementary Materials), so it is probable that the material remains in a contracted form during the catalysis. Indeed, after catalysis and ammonia TPD, there is no evidence for pore swelling in the PXRD (Supplementary Materials). Our measured BET surface area of 16 m<sup>2</sup> g<sup>−</sup><sup>1</sup> for MIL-88B(Fe,Sc) activated in vacuum at 100 ◦C corroborates this, showing the lack of porosity and a surface area as expected for micron-sized powders. This result is similar to that reported for other MIL-88B(Fe) materials: for example, Rahmani and Rahmani measured a BET surface area of 23 m<sup>2</sup> g<sup>−</sup><sup>1</sup> [65], while Vuong et al. found a surface area of 30 m<sup>2</sup> g<sup>−</sup><sup>1</sup> for a hydrated, mixed (Fe, Ni) form of the material with amino-functionalised linkers [66]. Previous work on dense (non-porous) scandium-organic frameworks has shown how surface Lewis acidity is sufficient for the acetalization of aldehydes or the cyanosilylation of carbonyl compounds [67,68]. Although Mitchell et al. found that MIL-88B(Sc) was an inferior Lewis acid catalyst compared to porous MIL-100(Sc) in a variety of organic transformations [41], which they attributed to its low surface area, in DMSO it is clear that the Lewis acidity is sufficient for the glucose isomerisation.
