Selective Electrocatalytic Oxidation of Biomass‐Derived 5‐Hydroxymethylfurfural to 2,5‐Diformylfuran: from Mechanistic Investigations to Catalyst Recovery

Abstract The catalytic transformation of bio‐derived compounds, specifically 5‐hydroxymethylfurfural (HMF), into value‐added chemicals may provide sustainable alternatives to crude oil and natural gas‐based products. HMF can be obtained from fructose and successfully converted to 2,5‐diformylfuran (DFF) by an environmentally friendly organic electrosynthesis performed in an ElectraSyn reactor, using cost‐effective and sustainable graphite (anode) and stainless‐steel (cathode) electrodes in an undivided cell, eliminating the need for conventional precious metal electrodes. In this work, the electrocatalysis of HMF is performed by using green solvents such as acetonitrile, γ‐valerolactone, as well as PolarClean, which is used in electrocatalysis for the first time. The reaction parameters and the synergistic effects of the TEMPO catalyst and 2,6‐lutidine base are explored both experimentally and through computation modeling. The molecular design and synthesis of a size‐enlarged C 3‐symmetric tris‐TEMPO catalyst are also performed to facilitate a sustainable reaction work‐up through nanofiltration. The obtained performance is then compared with those obtained by heterogeneous TEMPO alternatives recovered by using an external magnetic field and microfiltration. Results show that this new method of electrocatalytic oxidation of HMF to DFF can be achieved with excellent selectivity, good yield, and excellent catalyst recovery.


Introduction
Owing to the growing awareness of the inconvenient utilization of diminishing fossil resources,t he fast-rising levels of carbon dioxide emissions, and the ever-increasing demandi n energy,b iomass-based chemical platforms have gained much interest. In particular,t he utilization of agricultural wastes shows great promise. Catalytic transformation of lignocellulosic biomass into value-added chemical compounds could provide ar enewable, carbon-neutral feedstock platform that might be as ustainable alternative to the crude oil and natural gas based bulk chemical industry. [1] Within the furan family,5 -hydroxymethylfurfural (HMF) is a potentialC 6 carbohydrate-based buildingb lock and is attracting al ot of interest ( Figure 1). [2] Being accessible by the acidcatalyzed dehydration of hexoses, HMF is also an aturallyo ccurring substance, and its market, which is increasing rapidly worldwide, is expected to reach61millionUSD in 2024. [3] As aplatform chemical, HMF can be transformed into several high-value derivatives. [4] 2,5-Furandicarboxylic acid and its dimethyl ester are both promising monomers to produce furanbased polyesters as an alternative to the petrochemical-based polyethylene terephthalate (PET). [5] The hydrogenated diol de- The catalytic transformation of bio-derived compounds, specifically 5-hydroxymethylfurfural (HMF), into value-added chemicals may provide sustainable alternatives to crude oil andnatural gas-based products. HMF can be obtained from fructose and successfully converted to 2,5-diformylfuran (DFF) by an environmentally friendlyo rganic electrosynthesis performed in an ElectraSyn reactor, using cost-effective and sustainable graphite (anode) and stainless-steel (cathode) electrodes in an undivided cell, eliminating the need for conventionalp recious metal electrodes. In this work, the electrocatalysis of HMF is performed by using green solvents such as acetonitrile, g-valerolactone, as well as PolarClean, which is used in electrocataly-sis for the first time. The reaction parameters and the synergistic effects of the TEMPOc atalyst and2 ,6-lutidine base are explored both experimentally and through computationm odeling. The molecular design and synthesis of as ize-enlarged C 3symmetric tris-TEMPO catalystare also performed to facilitate a sustainable reactionw ork-up through nanofiltration. The obtained performancei st hen compared with those obtained by heterogeneous TEMPO alternatives recovered by using an external magnetic field and microfiltration. Resultss how that this new methodo fe lectrocatalytic oxidation of HMF to DFF can be achieved with excellent selectivity,good yield, and excellent catalystrecovery.
rivatives, 2,5-bis(hydroxymethyl)furanand 2,5-bis(hydroxymethyl)tetrahydrofuran, are both valuable polymer-building blocks for the synthesis of polyurethanes, aromatic resins,a nd polyesters. [6] 2,5-Diformylfuran( DFF), which contains two reactive aldehydeg roups, is ap articularly useful derivativeo fH MF (Scheme 1) with potential applicationsa sa ni ntermediatef or pharmaceuticals, [7] functional polymers, [8] fungicides, [9] macrocyclic ligands, [10] organic conductors, [11] and as ac rosslinking agent of poly(vinyla lcohol) forb attery separations. [12] This bis-(aldehyde) is usually synthesized by oxidation of the primary hydroxyl group of HMF,a nd owing to the reactive nature of the CHO, selectivity plays ak ey role in the efficiency of the production.T herefore, several homogeneous and heterogeneous metal-promoted (vanadium, manganese,a nd precious metals) oxidationp rocedures have been suggested for the synthesis of DFF. [13] Electrochemical platforms have the potential to provide an environmentally friendly solution fort he oxidation of sensitive compounds. Because of the multitude of adjustable reaction parameters such as electrode materials, electrolyte, solvent, currents trength, potential, the selectivity of the reactionc an be fine-tuned. Furthermore, by using renewable energy sources and recyclable catalyst/electrolyte systems, electroorganic methodologies could offer sustainable synthetic processes. [14] The scientific literature on the electrocatalytic oxidation of HMF to DFF is scarce (Table 1). Skowroń ski et al. performed a selectivee lectrooxidation with aP tanode in ab iphasic system, using acetic acid or inorganic salts as supporting electrolytes; [15] Cao et al. utilized aP tRu alloy to exploit the simultaneous generation of electricity on the cathode in am embraneelectrode reactor. [16] In addition to direct electrolysis, N-oxyl radicals are commonly used catalysts for the indirect oxidation of primary and secondary alcohols. [17] Particularly,2 ,2,6,6-tetramethylpiperidinyl-N-oxyl (TEMPO) and its derivativesa re commono xidants with industrial-and laboratory-scale applications. [18] Under electrochemical conditions, the formation of the active reactant from persistent organic radicals can be accomplished in the absence of chemical oxidants. [19,20] The catalyst-promoted electrooxidative synthesis of DFF in ab iphasic system, using 4acetamido-TEMPO and ar ecyclable NaHCO 3 (aq)/KI electrolyte, was demonstrated. [21] Figure 1. Annual number of publications related to HMF and DFF.Search engine:Web of Science;keywords:5-hydroxymethylfurfural and 2,5-diformylfuran; 16.10.2019. In the pursuit of sustainable chemical transformations, catalyst recovery plays ap ivotal factor for meeting ecological and economical demands. The recovery and reuse of TEMPOs have generally required ag reat variety of solid-supported heterogeneous and homogeneous organic supports. [22] Althought he recovery of insoluble catalysts is straightforward, the catalytic activity and selectivity may become impaired when anchored to solid carriers. On the contrary,h omogeneous catalysts could grant exceptionala ctivity and selectivity,b ut their inefficient recovery is ap roblem yet to be solved. [23] Organic solventn anofiltration (OSN) is as ustainable recycling technique for homogeneous catalysts. [24] Its scale-up and implementation in continuous processes are rathers traightforward, therefore feasiblef or industrial utilizations. As the efficiency of separation is largely dependent on the molecular weightg ap between the catalyst and the other solutes,s izeenlargement of small catalysts is generally required.C onsequently,h erein, we explore commerciala nd size-enlarged TEMPO catalystr ecovery by meanso fm agnetism, microfiltration, and nanofiltration (Scheme 1).
Recently,w eh ave demonstrated that the MCM-41-supported metal catalyst promoted the conversion of carbohydrates into HMF, [25] and herew er eport aT EMPO-mediated electrocatalytic oxidation methodf or the selective transformation of HMF into DFF (Scheme 1). The commercially availablec ompact electrolysis cell (IKA ElectraSyn 2.0) as reactor,g reen solvents (MeCN, g-valerolactone, Rhodiasolv PolarClean), and non-precious-metal-based electrodes (graphite, stainless steel) were selected. To the best of our knowledge,t his is the first report on using an on-precious-metal-based electrode for selective HMF conversion.W ithr ational catalystd esign, supported by quantum chemicals tudies, an ew homogeneous size-enlarged C 3symmetric tris-TEMPO derivative (Hub 1 )w as synthesized to facilitate the recovery of the catalystbynanofiltration. Acomparison of the recovery and catalytic performance of commercially availableT EMPO derivatives (SiliaCAT TEMPO, TurboBeads TEMPO) and the OSN compatible Hub 1 -TEMPO was performed.

Results and Discussion
Electrocatalytic oxidation of HMF to DFF Electrocatalytic oxidation can be performed directly when electron transfer (ET) between the substrate and the anode takes place at the electrode surface in ah eterogeneous manner.B y employing ac atalyst, the ET involving the substrate becomes a homogeneous process. The latter indirect method can mitigate over-oxidized side product formation and electrode passivation, which is essential ford eveloping as ustainable process. Therefore, the oxidation of HMF gained from fructose [25b] was investigated in ad irect process by using ag alvanostatic setup (current:1mA).
Graphite (anode) and stainless steel (cathode)w ere chosen to achieve cost-effective operation. [26] This arrangement improveds ustainability as the commonly used platinum electrodes were replaced (Table 1). The direct oxidationw ith the LiClO 4 electrolyte avoided product formation, whereas the addition of the 10 mol %T EMPO catalyste nabled ar eaction with am oderate yield of 28 %( Figure 2a). The addition of 2,6-lutidine as the base resulted in virtually complete consumption (more than 99 %) of the startingm aterial (HMF) withoutt he formation of undesired byproducts. To analyze whether the base alone or the combined effect of the catalyst-base pair causedt he increased yield, the electrochemical oxidation was also performed in the presenceo ft he base but without TEMPO. Althought he yield decreased to 76 %, it was still found to be highert han that of the TEMPO-catalyzed method (28 %). Consequently,t he synergistic effect of the catalysta nd the base were further investigated with computational methods ( Figure 2b).
The schematic pathways of hydride ion transfer fort he conversion of HMF to DFF both in the presence and absence of the base are showni nt he energy diagram ( Figure 2b). The presence of the base lowers the activation energy by 58 %. The base polarizes the OÀHb ond of HMF through hydrogen bonding. The hydride ion transfers from HMF to the catalyst in the transition state, which is followed by the product formation as ar esult of the completion of the hydride ion transfer.T he calculated activation energy for the formation of DFF is 61.57 kJ mol À1 .
After selecting the applicable catalyst/base system,t he parameters influencing the synthesis of DFF weree xplored. First, 11 conventionala nd alternative solvents were tested in theo xidationp rocess (Figure 3a). The categorization of the solvents was done based on green solvent selection guides. [27] Among these, solvents with dielectric constants higher than 40, such as ethylene carbonate (EC) and propylene carbonate( PC), water,a nd dimethyl sulfoxide (DMSO), provided lower yields and produced ac onsiderable amount of unidentified side products.O nt he contrary,t he use of solvents with dielectric constantslower than 40 resulted in excellent yields.T his observation suggests that solvents with higher dielectric constants might have unfavorable effects on the solvation or stability of the ionic species. This could result in higher cell resistance, and consequently,i ns ide product formation such as over-oxidation or reaction with the solvent during the ET.T he solvent effect on electrochemical processes is ac omplex matter andf urther investigations are needed in this field. Amongt he green solvents, the best results were achieved with g-valerolactone (GVL), PolarClean (methyl-5-(dimethylamino)-2-methyl-5-oxopentanoate), and acetonitrile (MeCN), as complete conversions were obtained without side reactions. Both GVL and Polar-Clean are emerging green solvents with the potential to pave the way towards ustainable electrolysis. [28] To the best of our knowledge,t his is the first application of PolarClean in electrocatalytic reactions.
The effecto fc urrent strength on the oxidation was examined ( Figure 3b). Increasing the electric current to 2a nd 3mA resultedi nn os ignificant change in the outcome of the reaction. However,a fter 8h of constant current electrolysis at 5mA, almost no product could be detected in the reaction mixture owing to accelerated side product formation.D eformation of the electrodes wasa lso observed (see the Supporting Information).
Owing to the heterogeneous nature of the electrochemical process,t he effects of both stirring rate (Figure 3c)a nd reac-tion temperature (Figure 3d)w ere investigated. Althoughh ighers tirring speed resulted in slightly faster product formation,n os ignificant change in yield was observed by increasing the temperature from room temperature to 40 8C. At am ore elevated temperature (60 8C), as malld ecrease in the yield was detected after 8h,p ossibly owing to the over-oxidation of the desired DFF.T he increase in the concentration of TEMPO to 20 and 30 mol %r esulted in virtually no change in the reaction. Moreover,areticulated vitreous carbon (RVC) electrode was also tested as the anode (instead of graphite), but despite its larger surface area, no significant change in the rate of the reaction was observed. Also, the delicate structure of the RVC electrode presenteda dditional difficulties in comparison to the standard graphite electrode. Refer to the Supporting Information for further details.

Catalyst design and recovery
Twoc ommercially available solid-supported TEMPO catalysts (TurboBeads:5 0nmd iameter and 15 m 2 g À1 , SiliaCAT: 1.2 mmd iametera nd approx. 500 m 2 g À1 ) were applied in the electrocatalytic oxidation of HMF under the optimized reactionc onditions (Figure 4). Both compounds are heterogeneous catalysts with the TEMPO units immobilizedo ni ron oxide cores and silica gel, respectively.T hese inert and resistant solid supports enablet he facile separation of the catalyst from the reaction mixture by using an external magnetic field and microfiltration, respectively.A moderately slower reaction rate was observed in comparison to the homogeneous TEMPO system ( Figure S9 in the Supporting Information). The TurboBeads provided full conversion after 16 h, whereas SiliaCATp rovided a good yield (93 %) in 20 h. The reactions for both heterogeneous catalysts can be described with pseudo-first-order kinetics (k TurboBeads = 0.1868 s À1 , k SiliaCAT = 0.1335 s À1 ), whereas no satisfactory correlation for the homogeneous TEMPO was observed.
To overcome the difficulties in recovering the native TEMPO and the slower reaction rate of the heterogeneous TEMPO derivatives, as ize-enlarged TEMPO for membrane recovery was designedi ns ilico. Catalysts with high molecular weight (M W ) exhibit high retention by nanofiltration membranes.
Owing to the small M W gap between TEMPO and the other reactionc omponents in the electrocatalytic oxidation, size-enlargemento fT EMPO was required to facilitatei ts recovery by organic solvent nanofiltration (OSN). Catalyst anchoring to soluble macromolecules [29] or small trifunctional hubs [30] is an efficient approachu sed in the recycling of high-value organocatalysts. The latter approach exploitst he C 3 -symmetric multifunctional hub to provide straightforward synthesis, facile characterization, and high catalytic unit to inactive moiety ratio. The hub, bond type, or bond length between the hub and the catalysta llows fine-tuning of the catalytic activity ande nables catalystrecovery ( Figure 5). Accordingly,e ights ize-enlarged TEMPO derivatives( Hub x -TEMPO)w ere considered for the oxidation of HMF ( Figure 5) by using the M062X/6-31G*l evel of the density functional theory (DFT) as implemented in the Gaussian software.T hree hubs, namely benzene, 1,3,5-triazine, and 1,3,5-triphenylbenzene, and five covalentb onds, namely ether,a mine, ester, amide, and 1,2,3-triazole, were studied. The triazine type Hub 3 -TEMPO has the smallest radius (8.3 ), whereas the Hub 8 -TEMPO with the triphenylbenzene unit has the largest (13.5 ). The size-enlargement resulted in an increasei nt he catalyst radius by as much as 286 %. The corresponding energy diagrams for the formation of DFF when using the different TEMPO derivatives are shown in Figure 5b.T he activation energy for the formation of DFF is lower for Hub 1-3 -and Hub 7 -TEMPOst han for the native TEMPO. Therefore, we conclude that the formation of DFF from HMF is kinetically more favorable when using these derivatives rather than the others, including the native TEMPO. In particular, Hub 2 -TEMPO,c losely followed by Hub 1 -TEMPO,s howedt he lowest activation and product energies of 27.73 and À92.80 kJ mol À1 ,r espectively. The synthesis of Hub 1 -TEMPO was found to be the most straightforward through the O-alkylation of 4-hydroxy-2,2,6,6tetramethylpiperidinyl-N-oxyl (1)w ith 1,3,5-tris(bromomethyl)benzene (2), which produced the catalyst with excellenty ield (93 %, Figure 6a). To confirm its structure by NMR spectroscopy (Figures S25 and S26 in the Supporting Information), Hub 1 -TEMPO was successfully reduced with l-ascorbica cid (Scheme S9 in the SupportingI nformation). Refer to the Supporting Information for the experimental protocol and spectra.
Electron paramagnetic resonance (EPR) spectroscopy showed ad ownshift in the g-value of the solvated native TEMPO( from 2.00641t o2 .00516) as ar esult of the size-enlargement (Figure 6b).  This phenomenonm eans that the free radical electrons are more looselybound to the nitrosyl oxide,which could enhance the catalytic activity.T hese findings are in line with the predictions of the DFT study.
Nonetheless, in comparison to the native TEMPO, the homogeneous Hub 1 -TEMPO added in equivalent mole percentage showedn os ignificant differences in yield or the progression of the reaction (Figure 4). Evenw hen the catalyst was used such that an equivalent amount of TEMPO units was present in the reactionmixture (one third the mole percentage in comparison that of the native TEMPO), virtually no change was observed in the catalytic performance. We can conclude that the size-enlargement did not adversely affect the catalytic performance. The GMT-oNF-1, NF030306, and DM300 membranes were screened to identify the most suitable membranef or the catalyst recovery by diafiltration ( Figure 7a). Based on the molecular weights, the rejection gap between the commercial TEMPO and the other solutes (approx. 50 %), as well as the absolute rejection of TEMPO (approx. 30-70 %) were not sufficiently large for successful diafiltration.O nt he contrary,t he rejection of Hub 1 -TEMPO was found to range between9 0% and 100 %f or all the membranes. In particular, DM300 fully retained the enlarged catalyst while still being able to effectively purge all other solutes, showingr ejection as low as 10-20 %. DM300 also demonstrated ah igh flux of 22 AE 0.4 Lm À2 h À1 ,w hich was 3.3 and 2.3 times highert han that of the GMT-oNF-1 and NF030306 membranes, respectively.C onsequently,D M300w as chosen for the catalyst recovery using diafiltration (Figure 7b). The concentration profile revealed that the solutes were completely purged out of the system within 10-12 diavolumes, and the catalyst purity reached 100 %. The highlighted area shows the results of the mathematical modeling for the catalyst purity when the other solutes showedr ejections between 10 %a nd 30 %, requiring1 0a nd 12 diavolumes to reach virtually 100 %c atalyst purity.T his result demonstrates the robustness of the proposed nanofiltration-based catalyst recovery.

Conclusions
Biomass-derived HMF was successfully converted to DFF with 78 %i solated yield and virtually1 00 %s electivity by utilizing the compactE lectraSyn reactor in ag alvanostatic setup in an undivided cell for environmentally friendly organic electrosynthesis. In comparison to the previous literature reports that  ChemSusChem 2020, 13,3127 -3136 www.chemsuschem.org 2020 The Authors. Publishedb yWiley-VCH Verlag GmbH &Co. KGaA, Weinheim employedp latinum as the electrode material, graphite (anode) and stainless steel (cathode) were chosen to achievecost-effective operation in this study.A mongt he green solvents tested, PolarClean was successfully used in electrocatalysis for the first time. The effects of current strength, stirring rate, temperature, catalystm olar ratio, electrode surfacea rea, and the roles of TEMPO and the lutidine base on the electrooxidation were explored both experimentally and through DFT modeling. Computer-aided modeling was used for size-enlarged catalyst design and structure optimization. The reaction pathways of the electrocatalytic conversion were determined, and the relative energy profiles of the native and designed catalysts were compared. Synergetic effectso fT EMPO and lutidine were observed, ensuring high yield and selectivity simultaneously.T he homogeneous size-enlarged C 3 -symmetric tris-TEMPO derivative was successfully recoveredb yu sing organic solvent nanofiltration.
The electrochemical experiments were carried out by using an IKA ElectraSyn 2.0 potentiostat equipped with either as ingle vial holder,o rasix-reaction carousel, or aG OGO module connected to an IKA KS 4000 ic ontrol shaker.T he reactions were conducted in constant current mode, without ar eference electrode. The electrodes and vials were purchased from IKA. The electrodes were washed multiple times with water,a nd acetone, and were rubbed dry with tissue paper before each use.
Infrared spectra were recorded with aB ruker Alpha-T FTIR spectrometer (s:s trong, m: medium, w: weak). Electron paramagnetic resonance spectroscopy was carried out in an EPR spectrometer (Xenon series from Bruker) at room temperature, and the unit was operated in the X-Band mode with am icrowave frequency of 9.4-9.8 GHz and am odulation frequency of 100 kHz. An ER 221 Bruker cell tube with an inner diameter of 3mma nd an outer diameter of 4mmw as used to load the samples. For solid state measurements, the samples were mixed with KBr powder to dilute the concentration. The sweep width was set at 600 Gw ith am odulation amplitude of 0.4 G. The radio frequency power was set to 0.6325 mW with power attenuation of 25 dB. For solvated state measurements, the samples were solvated with acetonitrile. The sweep width was set at 8000 Gw ith am odulation amplitude of 4G.T he radio frequency power was set to 0.6325 mW with power attenuation of 25 dB. NMR spectra were recorded either with aB ruker DRX-500 Avances pectrometer (at 500 MHz for 1 Ha nd at 125 MHz for 13 C spectra) or with aB ruker 300 Avances pectrometer (at 300 MHz for 1 Ha nd at 75.5 MHz for 13 Cs pectra), as specified for each compound. High-resolution mass measurements were performed with aT hermo Exactive plus EMR Orbitrap mass spectrometer,w hich was used with aT hermo Ultimate 3000 UHPLC with 100 %m ethanol as the mobile phase. Melting points were recorded with aB oetius micro-melting point apparatus, and the observations were not corrected. Silica gel 60 F 254 (Merck) plates were used for thin-layer chromatography (TLC) and the spots were visualized either by ultraviolet light (254 nm) or by staining with an acidic H 2 O/EtOH solution of 2,4-dinitrophenylhydrazine (DNP). Silica gel 60 (70-230 mesh, Merck) was used for column chromatography.T he ratios of the solvents for the eluents are given in terms of volume (mL mL À1 ). Yields (except for isolated yields) were determined based on the HPLC chromatograms. For the detailed calculation procedure, refer to the Supporting Information.

Particle size determination
The particle size of SiliaCATw as determined by means of dynamic light scattering using aZ etasizer Nanoseries instrument (Malvern Panalytical). The sample was dispersed in deionized water inside a glass cuvette cell with asquare aperture and measured immediately after the preparation of the dispersion by shaking. The measurement was carried out at 25 8Cw ith an equilibrium time of 120 s. Te nr uns, each of duration 10 s, were performed for six data collections. For further details, refer to the Supporting Information.
Synthetic procedure for the preparationofH MF from fructose HMF was prepared based on our previously described procedure [25b] with some minor modifications:i natypical run, the conversion of fructose into HMF was conducted in ag lass flask (500 mL) equipped with ac ondenser.T he deep eutectic solvent system was formed with fructose (20 g,0.11mol,1equiv) and ChCl (60 g,0.43 mol,4equiv). Then, HCl (0.2 mL, 37 %) was added as the catalyst. The flask was placed into an oil bath and heated (100 8C) with vigorous stirring. After the reaction was completed, the black mixture was dissolved in saturated NaCl solution (10 mL) and then extracted with ethyl acetate (5 30 mL). Anhydrous sodium carbonate (10 g) was added to the obtained organic solution and filtered to remove water and acid. Then, the organic solvent was removed with ar otary evaporator.T he concentrate was dissolved in acetone (50 mL) and further distilled to obtain 13.2 g( 95 %) HMF. The spectral data were fully consistent with those reported in the literature. [25b] General procedure for the electrochemical oxidation of HMF into DFF Without taking precautions to exclude air and moisture, the Elec-traSyn vial (5 mL) equipped with as tir bar was charged with HMF (31.5 mg, 0.25 mmol, 1.0 equiv), TEMPO (3.9 mg, 0.025 mmol, 0.1 equiv), 2,6-lutidine (29 mL, 0.25 mmol, 1.0 equiv), LiClO 4 (53.2 mg, 0.5 mmol, 2.0 equiv), and MeCN (5 mL). The ElectraSyn vial cap equipped with the anode (graphite) and cathode (stainless steel) was inserted into the mixture. The reaction mixture was electrolyzed at ac onstant current of 1mAf or 20 h. Then, the vial cap was removed, and the electrodes were rinsed with CH 2 Cl 2 (10 mL), which was combined with the reaction mixture. The crude mixture was concentrated under reduced pressure. The resulting mixture was taken up in CH 2 Cl 2 (25 mL) and washed three times with water (10 mL). The organic phase was dried over anhydrous MgSO 4 and concentrated in vacuo. The crude product was purified by preparative TLC (SiO 2 ,C H 2 Cl 2 /MeOH 40:1) to give DFF (48 mg, 78 %) as a white solid. The yield was recorded as the average of three parallel experiments (standard deviation: AE 2%). R f = 0.73 (SiO 2 ,C H 2 Cl 2 / MeOH 20:1, visualized by DNP);m .p. 106-109 8C( lit. [15] 108-109 8C); The spectral data are fully consistent with those reported in the literature. [15] 1, 3,2,6,oxymethyl)benzene (Hub 1 -TEMPO) 4-OH-TEMPO (1,5 06 mg, 2.94 mmol, 3.5 equiv) was dissolved in dry THF (1 mL) in ad ried round-bottomed flask under N 2 atmosphere. Next, NaH (60 %d ispersion in mineral oil, 176 mg, 4.40 mmol, 5.25 equiv) was added, and stirred at room temperature until the intensive gas formation stopped. Then, as olution of 1,3,5-tris(bromomethyl)benzene (2,3 00 mg, 0.84 mmol, 1equiv) in dry THF (1 mL) was added to the reaction mixture and stirred for 2d,d uring which precipitation was observed. After completion of the reaction, MeOH (3 mL) was added dropwise, followed by evaporation under reduced pressure. The remaining material was taken up in ethyl acetate (25 mL) and washed three times with water (10 mL). The organic phase was dried with anhydrous MgSO 4 and concentrated under reduced pressure. The crude product was purified by column chromatography with gradient elution (SiO 2 ,H ex/ EtOAc 1:1t o2 :3) to yield Hub 1 -TEMPO (493 mg, 93 %) as an orange solid. The structure of Hub 1 -TEMPO was confirmed by the NMR spectra of the N-OH derivative (S1). Refer to the Supporting Information for further details.

Organic solvent nanofiltration
The membrane separations were performed by using at ypical crossflow nanofiltration rig (Figure 8). AM ichael-Smith-Engineers gear pump was used for the recirculation of the retentate, and the speed was set at 1.2 Lmin À1 .T he commercial membranes were washed with acetonitrile and conditioned under ap ressure of 30 bar for 24 hp rior to rejection and flux measurements to ensure that the system reached as teady state. The solvent flux was determined by measuring the volume of the solvent permeating through the membrane within ag iven time for ac ertain surface area. The solute rejection was obtained from the ratio of the permeate and retentate concentrations of the solutes. The diafiltration of the crude reaction mixture was carried out at 30 bar using a Figure 8. Cross-flow nanofiltration apparatus for catalyst recovery. ChemSusChem 2020, 13,3127 -3136 www.chemsuschem.org 2020 The Authors. Publishedb yWiley-VCH Verlag GmbH &Co. KGaA, Weinheim DM300 membrane with an active area of 52 cm 2 .F resh acetonitrile was continuously fed into the vessel to compensate for the solvent volume leaving the system through the permeate stream, thereby maintaining ac onstant system volume. Samples of the permeate and retentate streams were periodically taken for analysis. The number of diavolumes, defined as the volume ratio of the permeate and retentate streams at ag iven time, was used to describe the progress of the filtration. The recovered catalyst was reused multiple times, and its characterization is shown in Figures S33 and  S34 in the Supporting Information.

Computational methods
All quantum chemical calculations for the conversion of HMF to DFF were performed with the Gaussian 09 package [31] and the ground-state geometries were optimized by using the hybrid meta-exchange-correlation functional M062X with the 6-31G* basis set. The transition states were analyzed by means of frequency calculation (single imaginary frequency). The polarizable continuum model was used for solvation.