Dehydration of Glucose to 5‐Hydroxymethylfurfural Using Nb‐doped Tungstite

Abstract Dehydration of glucose to 5‐hydroxymethylfurfural (HMF) remains a significant problem in the context of the valorization of lignocellulosic biomass. Hydrolysis of WCl6 and NbCl5 leads to precipitation of Nb‐containing tungstite (WO3⋅H2O) at low Nb content and mixtures of tungstite and niobic acid at higher Nb content. Tungstite is a promising catalyst for the dehydration of glucose to HMF. Compared with Nb2O5, fewer by‐products are formed because of the low Brønsted acidity of the (mixed) oxides. In water, an optimum yield of HMF was obtained for Nb–W oxides with low Nb content owing to balanced Lewis and Brønsted acidity. In THF/water, the strong Lewis acidity and weak Brønsted acidity caused the reaction to proceed through isomerization to fructose and dehydration of fructose to a partially dehydrated intermediate, which was identified by LC‐ESI‐MS. The addition of HCl to the reaction mixture resulted in rapid dehydration of this intermediate to HMF. The HMF yield obtained in this way was approximately 56 % for all tungstite catalysts. Density functional theory calculations show that the Lewis acid centers on the tungstite surface can isomerize glucose into fructose. Substitution of W by Nb lowers the overall activation barrier for glucose isomerization by stabilizing the deprotonated glucose adsorbate.


Introduction
Driven by growing environmental concerns about the negative impact of the combustion of finite fossil resources, significant efforts are currently underwayt od evelop routes to fuels and chemicals based on renewable feedstock.Lignocellulosic biomass is considered as one of the most promising sources of renewable carbon for the sustainable production of fuels and chemicals. [1]It is envisioned that the conversion processes in biorefineries will involve al imited number of key intermediates (platform molecules)w ith aw ide range of downstream applications. [2]The production of such platform molecules from cellulose, hemicellulose, and lignin needs to be performed with good efficiency.I nt his context,5 -hydroxymethylfurfural (HMF), which can be obtained from glucose( the main constituent of cellulose) is considered to be one of the most versatile plat-form chemicals.A ccordingly,t he development of efficient catalysts for glucosed ehydration has been well studied.Ac omplete review on this topic is available. [3]In general, to achieve ah igh yield of HMF,i somerizationo fg lucoset oi ts more reactive fructosei somer is required prior to dehydration of the sugar.A lthough the xylose isomerase enzyme is commercially used for the production of high-fructose syrups, the main reason to develop heterogeneousc atalysts is the limited compatibility of enzymes with the acidic conditions necessary for the subsequents ugar dehydration step.Both base and Lewis acid catalysts have been explored for this purpose. [3]Lewis acids are preferred because their activity is not affected by strong Brønsted acids. [4]7] In addition to this two-step approach, one-pot strategies to directly convert glucoset oH MF have gained attention recently.Asindicated above, it is generally assumed that this reaction proceeds via fructosea sa ni ntermediate, although ar ecent investigation by Noma et al. indicated that ad irect glucose dehydration pathway should also be considered. [8][11] The use of the CrCl 2 /1-ethyl-3-methylimidazolium chloride system under relatively mild conditions showed the promise of Lewis acids in ionic liquid solvents. [9]Water-tolerant Sn-Beta zeolite and aB rønsted acid catalyst, such as HCl, form an effective combination to obtain HMF in good yield from glucose, especially in biphasic systems. [5]Am echanism based on isomerization and dehydration was also proposedb yS t åhlberg and co-workers for the boric acid catalyzed glucose dehydration in ionic liq-Dehydration of glucoset o5 -hydroxymethylfurfural (HMF) remains as ignificant problemi nt he context of the valorization of lignocellulosic biomass.Hydrolysis of WCl 6 and NbCl 5 leads to precipitation of Nb-containing tungstite (WO 3 •H 2 O) at low Nb content and mixtures of tungstite and niobic acid at higher Nb content.Tungstite is ap romisingc atalyst for the dehydration of glucose to HMF.C ompared with Nb 2 O 5 ,f ewer by-products are formed because of the low Brønsted acidity of the (mixed)o xides.In water,a no ptimum yield of HMF was obtained for Nb-W oxides with low Nb content owing to balanced Lewis and Brønsted acidity.I nT HF/water,t he strong Lewis aciditya nd weakB rønsteda cidity caused the reactiont o proceedt hrough isomerization to fructosea nd dehydration of fructoset oapartially dehydrated intermediate, which was identified by LC-ESI-MS.The addition of HCl to the reaction mixture resulted in rapid dehydration of this intermediate to HMF.T he HMF yield obtainedi nt his way was approximately 56 %f or all tungstite catalysts.Densityf unctional theory calculationss how that the Lewis acid centers on the tungstite surface can isomerize glucosei nto fructose.Substitution of Wb y Nb lowerst he overall activation barrierf or glucosei somerization by stabilizing the deprotonated glucosea dsorbate.
uids. [12][15] To obtain ag ood HMF yield, the strong Brønsted acidity of niobic acid needs to be reduced, for example, by phosphation. [13]18] The detailso f the mechanism of glucose conversion to HMF have already been discussed for al ong time.[21][22] The aldose-ketose isomerization is thought to occur either by proton transfer or by intramolecular hydride transfer.Int he presence of ab ase, the aldose is deprotonated and the isomerization proceeds through as eries of enolatei ntermediates followed by re-protonation.In addition to fructose, this also yields mannoseast he product of epimerization.Lewis acids catalyzet he intramolecular hydride shift that transforms the open form of glucose to the open form of fructose. [20]Dehydrationo ff ructose is facile and removes three water molecules.Some of the intermediates have been identified. [23,24]  this work, we focus on tungstite as ap otentialc atalyst for the dehydration of glucose to HMF.T he structure of tungstite (WO 3 •H 2 O) is characterized by distorted octahedral units of tungstena toms coordinating to five oxygen atoms and aw ater molecule.The octahedra share four corners in the equatorial plane to form sheets.I ts surfacem ainly contains strong Lewis acid W 6 + sites.We introduced Nb in the synthesis to modify the acidic properties and achieveahigh HMF yield.As alient finding in the present study is that tungstite exhibits very weak Brønsted acidity, slowing down the complete dehydration of sugarst oH MF.A saconsequence, ap artially dehydrated reaction intermediate was isolated by LC-ESI-MS.The addition of HCl rapidlyc onverts this intermediate into HMF.This paper is organized as follows.In the first part, we present the characterization of the different oxidesa nd their performance in glucose conversion in water.T he optimum HMF yield is obtained at an intermediate Nb content as ar esult of balancedL ewis and Brønsted acidity.I nt he second part, density functional theory (DFT) calculations are used to demonstrate how Lewis acidic Wa nd Nb sites catalyze glucose isomerization.After exploring variousb iphasic solvent mixtures, in the final partw es how that the low Brønsted acidity in tetrahydrofuran (THF)/water mixtures limits the complete dehydration of glucoset oH MF.I nat wo-step approachi nvolving HCl to dehydrate the intermediates,t he HMF yield obtained from glucose by all of the tungstite catalysts is similar.T his study provides insight into the mechanism of glucosed ehydration and shows an ovel approach to dehydrate glucose to HMF startingf rom weakly Brønsted acidic tungstite.

Experimental Section Synthesis of materials
Mixed niobium-tungsten oxides, denoted by Nb x -WO 3 where x indicates the Nb/W ratio (x = 0.033, 0.1, 0.2, and 1), and tungsten oxide were prepared by stirring appropriate amounts of NbCl 5 and WCl 6 in water at room temperature.The light green precipitate that formed during stirring for 8h was filtrated and washed with deionized water until the filtrate had an eutral pH.The samples were finally dried overnight at 60 8C.P/WO 3 and P/Nb 0.2 -WO 3 were prepared by stirring as uspension of the parent oxide in as olution of H 3 PO 4 (1 g/200 mL) at room temperature for 2days.The samples were retrieved by filtration and then washed with deionized water repeatedly until the pH of the filtrate was 6.These samples were dried at 60 8Cf or 10 h.

Characterization
Powder X-ray diffraction (XRD) patterns were measured on aB ruker D4 Endeavor powder diffraction system using CuK a radiation.Nitrogen sorption was measured on aM icromeritics Tristar 3000 system in static measurement mode atÀ196 8C.The samples were outgassed at 120 8Cf or 3h prior to the sorption measurements.The Brunauer-Emmett-Teller (BET) equation was used to calculate the specific area from the adsorption data (p/p 0 = 0.05-0.25).FTIR spectroscopy of adsorbed CO was used to evaluate the acid properties of Nb x -WO 3 and P/Nb x -WO 3 .I nfrared spectra (1200-4000 cm À1 )w ere recorded in transmission mode in aB ruker Vertex V70v FTIR spectrometer equipped with ad euterated triglycine sulfate detector.T he catalyst was pressed into as elf-supporting wafer with ad iameter of 13 mm and then placed in ac ontrolled atmosphere transmission cell equipped with CaF 2 windows.Prior to CO adsorption, the sample was evacuated for 2h at 50 8Cw ith the pressure in the cell below 2 10 À6 mbar.T he sample was then cooled to À196 8Cb yf lowing liquid N 2 through ac apillary spiraled around the catalyst wafer.A ni nitial spectrum was recorded at this stage.CO was dosed through as ample loop connected to as ixway valve.In this manner,a ccurate doses of 0.04 mmol CO were administered to the cell.Each IR spectrum was recorded by accumulating 64 scans at ar esolution of 2cm À1 .D ifference spectra were obtained by subtracting the initial spectrum of the treated catalyst from the spectra obtained at increasing CO coverage.The CO stretch region of the various spectra was used to determine the density of the Lewis acid sites (LAS) and Brønsted acid sites (BAS).The extinction coefficient is assumed to be constant over the narrow frequency range of 2150-2200 cm À1 .T he value of the molar extinction coefficient was 2.7 cm/mmol. [25]This value and the intensity of the relevant band were used to calculate the number of LAS and BAS using the Lambert-Beer law [Eq.1]: in which N is the density of the vibrating species (mmol g À1 ), A is the intensity of the band (cm À1 ), e is the molar extinction coefficient of CO (cm mmol À1 )and 1 is the wafer thickness (mg cm À2 ).

Catalytic activitym easurements
Glucose conversion experiments were performed in ac losed, magnetically stirred glass reactor under autogenous pressure at 120 8C.

Full Papers
The reaction mixture consisted of 1mLo fa na queous 1wt% glucose solution in which 0.1 gc atalyst was suspended.After the reaction, the mixture was quenched in water and the liquid part was analyzed by aS himadzu HPLC system with evaporative light scattering detector (ELSD) and UV detectors.Glucose, fructose, and mannose were detected by ELSD using aP revail Carbohydrate ES (Grace) column for separation.For sugars, the mobile phase (1.0 mL min À1 )w as acetonitrile/water (70:30 v/v) and the column temperature was 50 8C.HMF was detected by aU Vd etector (320 nm) with aP athfinder PS C18 reversed phase column.The mobile phase (0.4 mL min À1 )i nt his case was methanol/water (20:80 v/v) and the column temperature was 30 8C.The lactic acid concentration was determined by using aP revail Organic Acid column in combination with aU Vd etector (220 nm).The mobile phase (0.5 mL min À1 )w as 25 mm potassium phosphate buffer with pH 2.5 and the column temperature was 45 8C.
Further analysis of the reaction mixtures was by LC-MS (Agilent 1200).The LC-MS was equipped with ad iode array detector and autosampler,d irectly coupled to an ion trap mass spectrometer (Agilent ion trap 6320) through an electrospray interface.AH ypersil C18-AR column was employed using acetonitrile (ACN) and water containing 0.1 %f ormic acid (water/FA) as the mobile phases, and was eluted according to the following gradient:0min, 98 %w ater/FA; 10 min, 80 %w ater/FA; 11 min, 100 %A CN;1 3min, 2% water/FA.The flow rate was 0.2 mL min À1 and the injection volume was 5 mL.The diode array detector recorded spectra from 200 to 550 nm.The MS was operated under ESI negative ionization mode using the following parameters:d ry temperature 350 8C, dry gas flow 9Lmin À1 ,n ebulizer gas pressure 40 psi, capillary voltage 3500 V.The instrument acquired data in the range of m/z 50-500.

DFT calculations
All DFT calculations were performed by using the Vienna ab initio simulation package (VASP 5.3). [26]The frozen-core projector augmented wave (PAW)a pproach was employed to describe the interactions between core and valence electrons. [27]The generalized gradient exchange-correlation Perdew-Burke-Ernzerhof (PBE) functional and plane-wave basis set with ac utoff energy of 500 eV were used. [28]AM onkhorst-Pack mesh of 3 3 3a nd 3 1 3 kpoints were used to sample the Brillouin zone for bulk and surface geometry optimization, respectively.T he on-site Coulomb correction for the W5ds tates was included with av alue of U eff = 3.0 eV. [29]The calculations were assumed to be converged when the forces on each atom were less than 0.05 eV À1 ,a nd the tolerance for energy convergence was set to 10 À5 eV.B oth the lattice vectors and atom positions were fully relaxed.The optimized lattice parameters for bulk orthorhombic WO The minimum reaction energy path and corresponding transition state were determined using the nudged elastic band method (NEB) with an improved tangent estimate. [30]The maximum energy geometry along the reaction path obtained with the NEB method was further optimized using aq uasi-Newton algorithm.In this step, only the adsorbate was relaxed.Frequency analysis of the stationary points was performed by means of the finite difference method as implemented in VASP.S mall displacements (0.02 ) were used to estimate the numerical Hessian matrix.The transition states were confirmed by the presence of as ingle imaginary frequency corresponding to the reaction path.All the calculated energies were further corrected for dispersion interactions by singlepoint energy calculations using the DFT-D3 approach developed by Grimme and co-workers. [31]sults and Discussion Characterization and glucose dehydrationinw ater  Full Papers mize the reactionc onditions for the dehydration of glucoset o HMF.G lucose isomerizationw as performed in singlep hase water at at emperature of 120 8Ca taglucosec oncentrationo f 10 mg mL À1 .A fter a8 3h reaction, the HMF yield was 34 %a t ag lucose conversion of 85 %.Higher initial glucose concentrations led to lower glucose conversion at nearly constant HMF selectivity (see Ta ble 2).Table 3s hows that the glucose conversion andt he HMF yield increasew ithr eaction time.The highest HMF selectivity of 40 %w as obtained after a3hr eaction.
We then evaluated the performance of the Nb x -WO 3 set and Nb 2 O 5 at optimized conditions (120 8C, initial glucose concentration of 10 mg mL À1 ,areactiont ime of 3h).The results are shown in Table 4.The low HMF yield of 16 %a tf ull glucose conversion for niobic acid is consistent with the work of Nakajima et al. [13] The Nb x -WO 3 samples afforded higher HMF yields under similarr eaction conditions.Although Nb 1 -WO 3 and Nb 0.2 -WO 3 gave the highest and nearly similarH MF yields of approximately 35 %, the HMF selectivity was highest for Nb 0.2 -WO 3 .Afurther decrease of the Nb content resulted in lower glucose conversion and HMF selectivity.T he reusability of the catalystw as also tested for Nb 0.2 -WO 3 .F or this purpose, the spent catalysts were first washed with deionized water and then dried at 60 8Cf or 6h.T hese recycle experiments demonstrate that high glucosec onversion with only as malll oss of the HMF yield was possible for at least four cycles (Figure 2).
Although Brønsted acids are sufficientt oc onvert fructoset o HMF,s electivec onversion of glucose usually requires isomerization to fructose prior to dehydration,f or which Lewis acids are effective. [5,33] ccordingly,w ec haracterized the Brønsted and Lewis acid properties of the Nb x -WO 3 and Nb 2 O 5 samples by IR spectroscopy of adsorbed CO. Figure 3s hows the CO stretching regiono ft he IR spectra.The spectra contain three bands at 2132 cm À1 ,2 160-2166 cm À1 ,a nd 2180-2185 cm À1 , which are assigned to physisorbed CO, CO adsorbed on BAS, and LAS, respectively.I ti si mmediatelyc lear that the amount of BAS and LAS strongly increase with Nb content.F or the Nb x -WO 3 samples with x < 1, the CO IR band attributedt oB AS is located at 2166 cm À1 ,w hereas this feature shifts to 2163 cm À1 for Nb 1 -WO 3 and to 2160 cm À1 for Nb 2 O 5 .I ti sl ikely that the band in the Nb 1 -WO 3 spectra is ac omposite owing to the presence of slightly acidic WÀOH and NbÀOH groups.It should be noted that the acidity of niobic acid increases in the presence of water.T herefore, it might be that we do not probe all the acid sites that are present in the catalyst under workingc onditions.The BAS and LAS densities in these samples were estimated according to known procedures [25] from the spectra after CO saturation and the resultsa re listed in Ta ble 1.The Nb 2 O 5 sample contains much more BAS and LAS than the Nb x -WO 3 andW O 3 samples.The concentration of BAS and LAS of Nb 1 -WO 3 are roughly half of the values of Nb 2 O 5 , whereas the other samples with much less Nb contain much less acid sites.The decrease in the total acidity with decreasing Nb content is much more pronounced than the decrease of the surface area.Thus, the changes in acidity can in part be attributed to the different chemical compositiono ft he surface.Isomorphous substitution of W 6 + by Nb 5 + in tungstite will replace W 6 + ÀH 2 Ob yN b 5 + ÀOH, whichc an explain the decrease of the concentrationo fL AS (cf. the decreasing ratio of LAS and BAS in Ta ble1).The NbÀOH groups are expected to be   [a] X sugar = sugarc onversion, Y HMF = HMFy ield, S HMF = HMFs electivity.
Full Papers slightly basic.T he weak BAS in these samples likely arise from hydroxyl groupsp resent at the defects ites such as crystal edges.T he samples may also contain niobic acid.The CO IR and textural data suggest that this is only the case for the Nb 1 -WO 3 sample.Ta king into account these acidity differences, we can attribute the lower glucose conversion and higherH MF yield of the tungstite samples as compared with Nb 2 O 5 to the de-creasedB rønsted acidity.T he highest HMF selectivity was obtained with the Nb 0.2 -WO 3 sample.Whereas the HMF selectivity is higherf or the Nb 0.2 -WO 3 sample, Nb 1 -WO 3 can convert more glucose.T he lower HMF selectivity for Nb 1 -WO 3 is owing to strong acid sites originating from niobic acid.These strong BAS catalyze side reactions including oligomerization to humins,a se videntf rom the reactionm ixtures turning yellow to brownafter aprolonged reaction, as well as retro-aldol reactions that decompose sugars.In all of the reactionm ixtures, as mall amount of fructose was observed with ay ield below 1%.T his demonstrates that fructosei sareaction intermediate and that the LAS in the Nb x -WO 3 samples can isomerize glucose to fructose.The BAS catalyzet he dehydration of fructose to HMF.W ea lso attempted to use Nb 0.2 -WO 3 as an isomerization catalyst by lowering the reaction temperature.However, at 100 8Ct he glucose conversion to HMF was less than 10 % and the fructose yield was below 1%.

DFT calculations on glucose isomerization
Because these preliminary data demonstrate that tungstite should be an active catalysti ng lucose isomerization,w eu sed DFT calculations to investigate possible reactionp athways for glucosei somerization on the stable (010) surface of ortho-rhombicW O 3 as am odel for the WO 3 •H 2 Os urface.Based on earlierw ork, [19,20] we explored am echanism involving three steps:( i) a-d-glucopyranose (Glu)r ing openingt ot he acyclic form of glucose( o-Glu), (ii)hydride shift from the C2 to the C1 atom in glucose, and (iii)ring closure of acyclicf ructose (o-Fru) to form a-d-fructofuranose( Fru).Glucose coordinates through its O1 atom sites to as urface tungsten site in ap lanar adsorption model (Figure 4).The ring opening of Glu is initiated by deprotonation of the O1H hydroxyl group, which involves the adjacent terminal Oa tomo fW 6 + =Oa sap rotona cceptor.This step generates an anionic Glu intermediate containing one negative chargel ocated at the O1 atom and ar educed WÀOH moiety in which the adjacent Wa tom has the 5 + oxidation state.The re-protonation of the O5 atom of Glu leads to opening of the glucoser ing.The proton in the resultingW ÀOH is then transferred to the O5 site.This step re-oxidizes the W 5 + À OH surface intermediate back to the W 6 + =Os tate.The activation barrier for the ring-opening step is only 34 kJ mol À1 .A fter this step, the isomerization reaction takesp lace.To facilitate the O2H deprotonation process, o-Glu has to change its coordination mode from O1ÀWc oordination to O2HÀW.Although this conformational change drastically disturbs the H-bonding interaction of the hydroxyl groups of o-Glu with the WO 3 surface, we predict that such reorientation will proceed with al ow barrieri na queous solution because the solventm olecules provide efficient stabilization through the hydrogenbondedn etwork.Deprotonation of O2H by W 6 + =Or esults in the anionic o-Glu intermediate andaW 5 + ÀOH group.This step is endothermic by 61 kJ mol À1 .T he isomerization step itself proceeds through the intramolecular H-shift from C2 to C1

Full Papers
with an activationb arriero f9 8kJmol À1 .T he terminal O1 atom of the anionic o-Fru intermediate is then protonated by ac oadsorbed H 2 Om olecule coordinating to an earby W 6 + =O( W 6 + =O•••H 2 O).Finally,f ructosei sp roduced by ring closure.As solvent effects in these polar reactions are important, [19] we also investigated the effect of additional water molecules on the O2H deprotonation and O1 protonation reactions.The activation barrierf or the H-shift reactioni nt he presence of more water decreased from 98 to 75 kJ mol À1 .W et hen investigated the influence of substituting Ww ith Nb on the WO 3 surface.This substitution lowerst he deprotonation energy of the O2H moiety to 15 kJ mol À1 ,w hereas the barrier for the H-shiftr eaction remains nearly unchanged( 91 kJ mol À1 ).Thus, the overall barrierf or the H-shift reaction is lowered to 106 kJ mol À1 as compared to the value of 132 kJ mol À1 computed for the pure WO 3 model surface.Qualitatively, the lower barrierf or the mixed oxide surface is consistentw ith the higherg lucosei somerization activity of the Nb x -WO 3 mixed oxidesincomparison to WO 3 .I ns ummary,t hese computational modeling results show that glucose can be isomerized in water on the Lewis acid surfaces ites of WO 3 and that Nb substitution on the WO 3 surfacef acilitates sugar deprotonation, thereby loweringt he overall activation barrierf or isomerization.

Effect of solvent on glucosedehydration
As it is well known that the use of other solvents and biphasic solvents ystemsc an significantly increase the selectivity of glucose dehydration, [5] we compared the performance of the optimum Nb 0.2 -WO 3 sample in different solvents.The resultsa re summarized in Table 5.In DMSO,t he HMF yield is only 15 %a t almostf ull conversion of glucose.T he low HMF selectivity is possibly caused by glucose condensation reactions in anhydrous DMSO. [35]In biphasic 1-butanol/water, [5] the HMF yield improved to 52 %a fter a3hr eaction compared with the experiments in water.T he increasedp erformance is owing to the extraction of HMF to the organic layer,w hich protects it from furthera cid-catalyzed side reactions in water such as the formation of humin.We also tested this catalysti namixture of THF/H 2 Oi na90:10 v/v ratio and found that with increasing reaction time the HMF yield increased only slowly.F igure 5 showst hat, despite high initial glucose conversion,t he HMF yield using Nb 0.2 -WO 3 was only 6% after 30 min.After a3hr eaction, glucosec onversionw as nearly complete at aH MF yield of 38 %.However,a fter a1 2h reaction the HMF yield was 62 %.a] Solvent [a] X glucose = glucose conversion, Y HMF = HMF yield.Full Papers reactionm ixture obtained after 3h reaction to 1.After further reactionf or 1h at 120 8C, the HMF yield increased to 58 % (Table 6).At 5a nd 10 times higherg lucose starting concentration, the HMF yields were still 53 %a nd 29 %, respectively,i f the experiment was performed in this two-step approach.In the next section, we discusst he use of phosphate-modified catalysts and LC-ESI-MS to identify the intermediate products.

Mechanistic considerations
In an attempt to gain an understanding of the role of the weak Brønsted acidity in the slow formation of HMF and the optimum conditions for the identification of possible reaction intermediates, we modified the acidity of Nb 0.2 -WO 3 and WO 3 using phosphoric acid.We included WO 3 at this stage because we hypothesized that the lowest HMF yield obtainedf or the WO 3 sample could be because of to the low acidity of this sample.Characterization of the phosphated samples by CO IR spectroscopy (Table 6a nd Figure 6) confirmed that the ratio of LAS to BAS of these phosphated samples was highert han those of the parentm aterials.This explainst he lower HMF yields of P/Nb 0.2 -WO 3 and P/WO 3 comparedt ot he parents amples.For Nb 0.2 -WO 3 ,t he HMF yield decreased from 38 %t o 30 %.[18] In our case, the decrease of the Brønsted acidity upon phosphationo fN b 0.2 -WO 3 and WO 3 results in decreased HMF yield, presumably because, different from the other materials, the rate-controlling step is dehydration of ar eaction intermediate.Consistent with this, we found that the addition of HCl to bring the pH of the solution to 1f ollowed by furtherr eactionf or 1h at the same temperature resulted in as teep increase of the HMF yield to 58 %.These resultsu nderpin our assumption that the reaction mixtures contain ar elativelys table reaction intermediate that can be dehydrated to HMF by HCl.As these data imply that the Brønsted acidity of the tungstite samples is too weakt oc ompletely dehydrate glucoset oH MF,a nd fructosei sa ssumed to be an intermediate in the overall reaction, we also studied the dehydration of fructose by WO 3 and Nb 0.2 -WO 3 before and after phosphation (Table 6).Fructose conversion was complete after 2h for all samples.This observation, together with the observation that phosphation decreased the HMF yield, suggests that similarr eaction intermediates are formed duringg lucose and fructose dehydration.This result is also consistent with the low fructose yields observedd uring glucosed ehydration, that is, isomerization of glucose to fructosea nd dehydration of fructose to the unknown intermediate are much faster than the further dehydration of the intermediate to HMF.T he fructosec onversion results also show that the BAS of (phosphated) tungstite are already sufficiently strong to initiate the dehydration of fructose.We also verified that the HMF yield during fructose conversion is very low if H 3 PO 4 is used as ac atalyst,  [b] 994 5.9 -3 100 58 100 [c] 53 [c] 99 [d]    Full Papers which shows that the phosphate groups do not contribute to the conversion of fructose.On the contrary,areasonable HMF yield was obtained from fructosew ith HCl.
As the influence of phosphationw as largest for WO 3 ,w e used the P/WO 3 sample to study glucose dehydration in more detail.As the reaction proceeded, the HMF selectivity increasedm uch slowert han the glucosec onversion (Table 6).Fructose yields were below 1% in this experiment.After 5h, the HMF yield was less than 10 %.In anothere xperiment, we removed the P/WO 3 sample from the reactionm ixture after 1h by filtration and adjusted the pH of the filtrate to 1b yH Cl.This led to af ast increase of the HMF yield to 50 %a fter a3h reaction.We verified that mixingt he glucoses olution with an equivalent amount of HCl didn ot yield HMF,w hich is the expected result.These data further support the conclusion that dehydration of some reaction intermediates is slowed down by the weak acidity of the catalyst.We then reactedg lucose in aT HF/water mixture in the presenceo ft he Nb x -WO 3 with x < 1( the samples which did not contain niobic acid) for 3h followed by adjustment of the pH to 1a nd continued the reaction for 1h at the same temperature.In all cases, the HMF yield was very similar at 56 % AE 2% (Figure 7).Thus, we infer that the earlier differences in the HMF yields observed for the Nb x -WO 3 samples are, in large part, relatedt oa cidity differences that lead to varying amounts of the unknown reaction intermediates dehydrating to HMF.I nc ontrastt oo ther catalyst systems, in which isomerization limits the reactionr ate, the dehydration of the relativelys table reaction intermediates is most likely the rate-limiting step on tungstite.This opens the possibility for at wo-step glucose conversion strategy consisting of isomerization on LAS and partial dehydration under weakly acidic conditions on BAS followed by dehydration of the intermediates by am ineral acid.This approach results in HMF yields of close to 60 %, which is comparable to other transition metal oxide systems, [8,[13][14][15][16][17][18] yet lower than resultso btained with Sn-Beta/HCl in THF/water. [5]Thea dvantage of the use of tungstite catalysts is that they can operate at lower temperature and the catalyst is much easier to preparet han Sn-Beta.
Finally, we attempted to identify the nature of the unknown reactioni ntermediate(s).For this purpose, we analyzed the reaction mixtures after 3husing P/WO 3 as the catalystu sing LC-ESI-MS( see the SupportingI nformation).We observedt he presence of HMF, two compounds with molecular masses of 144 gmol À1 and 158 gmol À1 ,u nconverted sugars as monomers and oligomers (i.e.,w ith other sugars, HMF,a nd the 144 gmol À1 compound), and 1,5-anhydroglucitol.I nt he cyclic reaction mechanismst hat are usually considered for fructose dehydration, the reaction of ketofuranose starts with dehydration of the hemiacetal at the C2 position, followed by two consecutive b-dehydrations leadingt oH MF (Scheme 1).Both of these intermediates (compound 1a nd compound 2) have earlier been identified by 13 CNMR. [23,24] he ESI-MS spectra suggest that the 144 gmol À1 compound is the product of the first two dehydration steps of fructose (compound 2i nS cheme 1).This would be (4R,5R)-4-hydroxy-5-hydroxymethyl-4,5-dihydrofuran-2-carbaldehyde, which has been identified as an intermediate in the dehydration of fructose to HMF in DMSO. [23][36][37] It is likely that this dehydrogenated product is as ide-producta ssociated with the stoichiometric oxidationofthe alcohol groups by the tungsten oxide catalyst into the stable dehydrogenated side-product.In keepingw ith this, we observed the yellow WO 3 sample turn blue, indicative of the reduction of W 6 + to W 5 + .W es peculate that this product is relatively stable, whereas com-pound2will be rapidly dehydrated to HMF by HCl.

Conclusions
Mixed niobium-tungsten oxidesw ere obtained by precipitation from aqueous solutionso fN bCl 5 and WCl 6 .T he samples had the tungstite structure and at ah igh Nb content also contained niobic acid.The Brønsted and Lewis acidic properties of these oxides were compared to those of Nb 2 O 5 by CO IR spectroscopy.Mixed Nb x -WO 3 and WO 3 catalyze the dehydration of glucoset oH MF in water with good yield.The mixed Nb-W oxides have much lower Brønsted acidity than Nb 2 O 5 and consequently,f ewer by-productsw ere formed during glucosec on-

Full Papers
version.The optimum HMF yield was obtained for mixed oxides with balanced Lewis and Brønsted acidity.Although glucose is rapidlyc onverted by tungstite in THF/water,t he HMF product is formed slowly.I somerization and dehydration occur at ah ighr ate on the LAS and the weakB AS of WO 3 and tungsten-richm ixed oxidesH owever,t he last dehydration step to HMF requires stronger Brønsted acidity.A ni ntermediate with am olecular mass of 144 gmol À1 ,l ikely (4R,5R)-4-hydroxy-5-hydroxymethyl-4,5-dihydrofuran-2-carbaldehyde, waso bserved by LC-ESI-MS.When HCl is added to the resultingr eaction mixture, the HMF yield is about 56 %i rrespectiveo ft he Nb content of the tungstite samples, confirming the slow dehydration of the partially dehydrated reaction intermediate.DFT calculations confirmt hat the tungstite surfacec an adsorb glucose, abstract ap rotonf rom glucose, and catalyze the hydrides hift to fructose.Nb substitution in the tungstite surfaces tructure stabilizes the deprotonated glucosei ntermediate and, accordingly,l owers the overall activation barrier for glucosei somerization.The present study suggestsanovel strategy of combining Lewis acidic transition metal oxides with weak Brønsted acid sites to catalyze isomerization and dehydration to relatively stable intermediates followedb yf urtherd ehydration by am ineral acid to obtain HMF.
3 •H 2 Ow ere found to be a = 5.39 , b = 11.03,a nd c = 5.19 .T he WO 3 •H 2 O( 010) surface model was employed to investigate the adsorption and isomerization of glucose by surface Lewis acid sites.A2 1 2supercell containing two layers of WO 3 •H 2 Oa nd av acuum slab of 15 along the (010) direction was employed.Each layer consisted of eight inplane WO 2 units as well as four top and four bottom out-of-plane oxygen atoms and H 2 Om olecules.Only the surface layer of WO 3 •H 2 Oa nd the adsorbate were fully relaxed, whereas the other atoms and lattice parameters were fixed during the surface calculation.The closed cell electronic configuration of the WO 3 •H 2 Ow as determined as the electronic ground state.

Figure 4 .
Figure 4. Reactionenergydiagramfor the isomerization of glucose on the (010) surface of WO 3 •H 2 O.For clarity, only the first coordination sphere of the central cation is shown.The central cation is W 6 + or Nb 5 + (all energies in kJ mol À1 ).
[a] X glucose = glucose conversion, Y HMF = HMF yield, N BAS = number of BAS, N LAS = number of LAS.[b] pH 1.0.[c] 50 mg Glucose.[d] 100 mg Glucosea dded.[e] After 2hreaction, the catalystw as removed and HCl was added and the reaction was continued for 3h.

Figure 7 .
Figure 7. Catalyticperformance for Nb x -WO 3 after 3hconversion of as olution of glucose in THF/H 2 O( 90:10, v/v) by the combination of the catalysts and HCl (the solvent pH is 1).The reaction temperature was 120 8C(red:g lucose conversion, blue:gray:H MF yield).

Table 1 .
Acid site densities of Nb 2 O 5, Nb x -WO 3 and WO 3 determined by IR spectroscopy of adsorbedC O.[a] [a] SA = surfacearea, N BAS = number of BAS, N LAS = number of LAS.