Carbohydrate Research 346 (2011) 664–672 Contents lists available at ScienceDirect Carbohydrate Research journal homepage: www.elsevier.com/locate/carres Converting fructose to 5-hydroxymethylfurfural: a quantum mechanics/molecular mechanics study of the mechanism and energetics Stavros Caratzoulas ⇑, Dionisios G. Vlachos Catalysis Center for Energy Innovation and Department of Chemical Engineering, University of Delaware, Newark, DE 19716, USA a r t i c l e i n f o Article history: Received 11 October 2010 Received in revised form 14 January 2011 Accepted 26 January 2011 Available online 4 February 2011 Keywords: Fructose dehydration mechanism 5-Hydroxymethylfurfural QM/MM molecular dynamics PM3 Free-energy calculations a b s t r a c t We studied the energetics of the closed-ring mechanism of the acid-catalysed dehydration of D-fructose to 5-hydroxymethylfurfural (HMF) by carrying out canonical ensemble free-energy calculations using bias-sampling, hybrid Quantum Mechanics/Molecular Mechanics Molecular Dynamics simulations with explicit water solvent at 363 K. The quantum mechanical calculations are performed at the PM3 theory level. We find that the reaction proceeds via intramolecular proton and hydride transfers. Solvent dynamics effects are analysed, and we show that the activation energy for the hydride transfers is due to reorganization of the polar solvent environment. We also find that in some instances intramolecular proton transfer is facilitated by mediating water, whereas in others the presence of quantum mechanical water has no effect. From a micro-kinetic point of view, we find that the rate-determining step of the reaction involves a hydride transfer prior to the third dehydration step, requiring an activation free energy of 31.8 kcal/mol, and the respective rate is found in good agreement with reported experimental values in zeolites. Thermodynamically, the reaction is exothermic by DF ¼ 20:5 kcal=mol. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction The diminishing availability of fossil resources to produce energy and much needed chemical materials has spurred a burgeoning research activity to discover and develop technologies that would involve renewable energy sources. Abundant biomass, as the only carbon-containing, renewable, primary energy carrier, promises to serve as a sustainable alternative that could supply valuable intermediates to the chemical industry. In particular, the class of carbohydrates—the prominent compound in biomass—possess a remarkable potential to act as a future resource. Nature produces a vast amount of 170 billion tonnes of biomass per year by photosynthesis, 75% of which can be assigned to the class of carbohydrates. Surprisingly, only 3–4% of these compounds are used in the food- and non-food sector.1 The economic viability of biomass-based processes depends critically on selectivity, a parameter that requires tight optimization. The many challenges involved in such an endeavour stem from the fact that carbohydrates are highly functionalized molecules. Their high content in oxygenated groups is a significant drawback for their conversion to fuels. On the other hand, the high functionality they possess is also an advantage, as the selective removal of some of these groups and the modification of others can lead to various value-added chemicals. The challenge is to develop cost-effective methods to ⇑ Corresponding author. E-mail address: [email protected] (S. Caratzoulas). 0008-6215/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.carres.2011.01.029 control the functionality in the final product.2 In this context, furan derivatives, such as 5-hydroxymethylfurfural (HMF), can be produced by the acid-catalysed dehydration of hexoses. HMF and its 2,5-disubstituted furan derivatives can replace key building-block molecules, currently derived from petrochemicals, in the production of plastics and fine chemicals.3,4 The acid-catalysed dehydration of D-glucose and D-fructose to HMF in aqueous media has received considerable attention over the years and more so the last decade. The production of HMF from glucose or fructose is typical for the selectivity and yield difficulties one encounters in the conversion of highly functionalized sugars, impeding high-volume production of HMF due to high costs. The dehydration rate of glucose is about 40 times lower than that of fructose, and much lower is the product yield. Kinetics studies by Kuster and co-workers show that the initial glucose or fructose concentration, the water concentration, the acidity of the aqueous medium or the presence of weak-acid anions (functioning as base catalysts) are all factors that influence the conversion of the sugars to HMF, the HMF selectivity and the rate at which HMF degrades by hydrolysis to levulinic and formic acids.5–7 At pH >3.9, no formation of HMF takes place, and at pH >2.7 no formation of levulinic acid occurs.7 Water displacement (by polyethylene glycol or other co-solvents) is also very beneficial to the formation of HMF from fructose because it accelerates its formation and retards its hydrolysis.6 Kuster reports only low HMF selectivity in environmentally benign solvents (water) and high selectivity in environmentally questionable ones [dimethyl sulfoxide, (DMSO)];8 DMSO is also a high-boiling point solvent and economic arguments can be made S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 against its use, too. Nevertheless, the case of DMSO is intriguing, as it has been suggested that DMSO preferentially stabilizes the furanoid tautomer of fructose (five-membered ring) to the pyranoid form (six-membered ring), implying that the dehydration of fructofuranose is more HMF selective. Amarasekara et al. have carried out mechanistic studies on the conversion of fructose to HMF in pure DMSO solvent, and by establishing the presence of a key intermediate, (4R,5R)-4-hydroxy-5-hydroxymethyl-4,5-dihydrofuran-2-carbaldehyde, have suggested that the DMSO catalyses the reaction.9 Antal and Mok have performed the reaction in suband supercritical water and achieved only unsatisfactory yields of HMF.10 Bicker et al. have used acetone–water mixtures as reaction media under sub- and supercritical conditions and reported 77% HMF selectivity and 99% fructose conversion. Interestingly, their kinetics and NMR data suggest that the dehydration of fructose to HMF seems to be most selective when the carbohydrate molecule is in its furanoid form. They have rationalized this finding on the grounds that the acetone molecule is similar to DMSO, thus suggesting that it favours the furanose tautomer.11 Román-Leshkov et al. have used DMSO or poly(1-vinyl-2-pyrrolidinone) (PVP) as ‘phase modifiers’ in a biphasic process for the acid-catalysed conversion of fructose to HMF. By managing to suppress undesired side reactions, they have achieved 80% HMF selectivity at 90% fructose conversion.12 It is unclear, however, how these modifiers function, from a mechanistic point of view. Despite the significant progress that has been reported recently, overcoming selectivity and yield difficulties related to the high functionalization of sugars requires tight optimization, before existing or even new processes become economically viable. This requires extensive microkinetic modelling and thus understanding of the fundamental chemistry, that is, of the reaction mechanisms and rates. In this article we focus on D-fructose and study the multistep mechanism and energetics of the acid-catalysed dehydration reaction to form HMF. The higher selectivity and yield compared to the direct dehydration of glucose as well as the development of new catalysts13 for the efficient and cost-effective production of fructose from glucose by isomerization make it an important reaction in the conversion of biomass to value-added chemicals. However, accurate energetics for the selective dehydration of fructose, required for process optimization and to design appropriate catalysts have not been reported. Modelling the solvent explicitly (to account for solvation effects), and the reacting system Quantum Mechanically (at the PM3 theory level) we employ hybrid Quantum Mechanics/Molecular Mechanics (QM/MM) Molecular Dynamics simulations to calculate, for the first time, reaction and activation free energies for every elementary step of the reaction. Details about the calculations are given in Section 2. This work also demonstrates that when investigating the mechanism of a reaction in solution it does not suffice to consider only the changes in the quantum mechanical energy of the reacting system, even if implicit or explicit solvent models are employed—one must examine the free-energy changes along the reaction coordinate monitoring the progress of the reaction. 2. Methods We study the energetics of the proposed mechanism by carrying out free-energy molecular dynamics calculations for each elementary step using a hybrid QM/MM Hamiltonian. The calculations were carried out in explicit water solvent modelled with the SPC/E molecular mechanics force field. As the mechanism involves cleaving and forming of chemical bonds, the fructose molecule was treated quantum mechanically at the semi-empirical PM3 theory level, as implemented in the AMBER 9 software package.14 Water molecules participating in the mechanism were also 665 included in the QM region and treated at the PM3 level. All the simulations were carried out in the canonical ensemble at the rather moderate temperature of T = 363 K. For the thermostat we used Langevin dynamics with collision frequency 2 ps1. The integration time step in all the simulations was set to Dt ¼ 1 fs. The ability of the PM3 method to model carbohydrate molecules was more recently demonstrated by Wu et al. who employed it to study the isomerization of fructose to glucose catalysed by a phosphoglucose isomerase.15 Pomata et al. have recently employed the SPC/E water model to study the hydrogen bond network and dynamics of fructose aqueous solutions up to 70 wt %.16 The free-energy calculations for every step of the mechanism were carried out using bias sampling along a properly chosen reaction coordinate. For example, in an elementary step involving proton or hydride transfer, we define the reaction coordinate as the distance between the transferring atom and the donor or acceptor atoms. For the bias potential we used the parabolic form V b ðnÞ ¼ ð1=2Þ kb ðn n0 Þ2 , where n is the reaction coordinate and a function of the atomic coordinates of the system. For the bias spring constant, kb, we used a range of values, depending on the anticipated energy barriers for the various elementary steps of the mechanism. For each elementary step, profiling of the free-energy change along a reaction coordinate entails a series of QM/MM MD simulations. These were performed according to the following protocol: (i) preparation of the quantum mechanical system and solvation—on average the simulation box contained 650 water molecules; (ii) optimization to remove accidental overlaps between atoms due to solvation; (iii) equilibration in the NpT ensemble for 200 ps at p = 1 atm and T = 363 K to fix the density of the medium (i.e., the simulation box size); (iv) another round of equilibration in the NVT ensemble for 200 ps at T = 363 K; (v) discretization of the reaction coordinate at a set of equidistant points fn0i gNi¼1 which define the centre of the bias potential V b ðnÞ and thus a sequence of N simulation ‘windows’ for biased sampling. In each simulation window, the system was propagated for 700 ps under the bias Hamiltonian before switching over to the next one. Of the 700 ps of propagation time, 200 ps were used for equilibration and 500 ps were used for the observation period, during which we saved configurations every 10 fs for post-processing. Care was exercised so that the bias energy at the centre of a simulation window does not correspond to energy higher than 2–3 kBT of the neighbouring bias potential, where kB is Boltzmann’s constant. Put differently, to ensure reversibility, the points fn0i gNi¼1 were chosen so that the domains of configuration space that were sampled in neighbouring simulation windows overlap. This requirement dictated the length increment between the points n0i , which was never larger that 0.1 Å. For example, scanning along strong C–H bonds with significant barriers to dissociation, required a rather aggressive bias potential with kb ¼ 200 kcal=mol A2 . A choice of Dn0 ¼ 0:1 A gives DV b ¼ 1 kcal=mol which at T ¼ 363 K is equivalent to 1:4 kB T, ensuring that thermal fluctuations provide the necessary overlap between neighbouring simulation windows. In a simulation ‘window’, we can calculate the biased marginal probability density, pb ðnÞ, in the reaction coordinate, n, that monitors the progress of the reaction in an elementary step, and from it the biased free-energy change. The unbiased free-energy change along n is given by DF ¼ kBTln pðnÞ , where pðnÞ is the unbiased marginal probability density. The latter is obtained from pb ðnÞ by the formula pðnÞ ¼ pb ðnÞexp½ðV b ðnÞ f Þ=kB T, where f is the free energy required to insert the bias potential V b into the system and depends on the simulation window. The free-energy profile for the elementary step is subsequently constructed by combining the unbiased probabilities across all simulation ‘windows’, using the weighted-histogram analysis method.17,18 Upon locating the freeenergy barrier along a reaction coordinate, and in order to obtain 666 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 accurate activation energies, we performed further sampling in the vicinity of the barrier, by adding simulation ‘windows’, and then re-calculated the free-energy profile in the augmented sample space. Typically, each simulation ‘window’ required about 5.5 h of CPU time for equilibration and observation; and a full energy profile for an elementary step required 25–30 ‘windows’. The unbiased expectation of a dynamical variable, say K, as a function of the reaction coordinate (i.e., the unbiased conditional expectation) is calculated by R hKiðnÞ ¼ dqKðqÞdðn nðqÞÞeðV b ðnÞf Þ=kB T eHb =kB T R dqdðn nðqÞÞeðV b ðnÞf Þ=kB T eHb =kB T ð1Þ where q collectively denotes the coordinates of the atoms in the system and Hb is the Hamiltonian of the biased system. 3. Results and discussion 3.1. Mechanism and energetics In the following, we label the fructose carbon atoms according to the convention depicted in Figure 1; to the oxygen atoms of the OH groups we assign the same label as the carbon atom to which they are attached, viz., the oxygen atom on the anomeric carbon (C2) is labelled O2, etc. We are investigating the energetics of the closed-ring mechanism of the acid-catalysed dehydration, whereby the fructofuranose ring remains intact. According to Antal and Mok,10,19 the preponderance data from chemical kinetics studies favour the closed-ring over the open-ring mechanism. Recent gas-phase electronic structure calculations of the acid-catalysed dehydration of xylose and glucose show that open-ring pathways to HMF are very unlikely as they involve very high activation barriers.20 The outcome of the QM/MM MD studies presented in this paper is the mechanism that we show in Figure 2. In the following we analyse its energetics. 3.1.1. Protonation and first dehydration We model the acid catalyst by protonating the OH groups of the fructose ring. Protonation of each of the three OH groups was considered separately and the free-energy change for the formation of the respective oxonium ion was calculated. As a free proton cannot exist in solution, in Figure 3a we show the free-energy change for the transfer of a proton from a hydronium, H3O+, to a fructose-ring OH as a function of the distance between the proton and the accepting oxygen atom. In this calculation the QM region consisted of the fructose molecule and the hydronium ion. We find that the protonation is an activated process, requiring activation free energy, DF z ; of 17:7 kB T (12.8 kcal/mol) for the OH on C2, 20.1 kBT Figure 1. Atom labelling convention for fructose. (14.5 kcal/mol) for the one on C3 and 18.9 kBT (13.6 kcal/mol) for that on C4. All three curves have a global minimum at 1.75 Å, which corresponds to the formation of a hydrogen bond between the OH group and the approaching hydronium ion. The reactions are endergonic, with that of the C2–OH being less so by about 4 kcal/mol. So, from a kinetic point of view, protonation of either one of the three ring OH groups is equally likely, with the protonation of the C2–OH being slightly favoured from a thermodynamic point of view. Qian et al. have carried out Car–Parrinello MD simulations for the dehydration of D-glucose and D-xylose at 500 K. They have proposed that the protonation of the OH group is probably the ratelimiting step.21 They predicate that conclusion on the observation that in the course of the MD trajectory (5 ps long), the proton, which was initially attached to one of the ring OH groups, transferred back to a water molecule in less that 100 fs, that is, quite rapidly. Transfer of the proton back to a water molecule is not unlikely, according to our calculations, as the barrier to de-protonation is in the range of 8–10 kBT at 363 K, with the upper bound corresponding to the C2–OH profile (see Fig. 3a). However, we find that the Classical Transition-State Theory (CLTST) lifetime of the protonated state is significantly longer in the case of D-fructose: 3 ns at 363 K, and 100 ps at the elevated temperature of 500 K; if one includes friction corrections, these numbers should go up.22,23 (Although for hydrogen transfer quantum mechanical corrections are important, at this point we content ourselves with the classical estimates as the reaction temperatures under consideration are high.24) As we shall see in the following, we do not find the protonation to be the rate-determining step of the acid-catalysed dehydration of fructose. The free-energy change for the removal of the protonated OH is shown in Figure 3b, for each of the three OH groups. The potential of mean force (pmf) is graphed as a function of the length of the cleaving bond. Removal of the protonated OH that is attached to the C2 carbon can take place in a facile manner, with an activation free energy of just 8.3 kBT (or 6 kcal/mol) and results in the formation of a stable intermediate, (30 ). The first dehydration is mildly exergonic with DF ¼ 3:1 kB T (2.2 kcal/mol). The water break-off is followed by charge transfer from the ring oxygen, O5, to the C2 carbon and the formation of a double bond between them, of average length 1.25 Å. Because of the sp2 hybridization of the C2 carbon, the O5–C2 and C1–O1 bonds assume a co-planar conformation. The effects of solvations are significant; they are analysed in Section 3.2. In comparison, initiation of the reaction by removal of the (protonated) OH on the C3 carbon is less likely, as the free-energy barrier is equal to 43 kBT (or 31 kcal/mol), which implies that the CLTST rate constant is 10-6 s1, viz., slower by fifteen orders of magnitude. Removal of the OH on C4 has a lower activation free energy, about 34 kBT (or 24.5 kcal/mol), but still an order of magnitude greater than the free-energy barrier for removal of the OH on C2 (8.3 kBT). Further dehydration of the resulting intermediate results in either ring opening or fragmentation and formation of very energetic free radicals. This is a parasitic pathway, but one that is surely initiated much slower (by twelve orders of magnitude) than the productive one. 3.1.2. Second dehydration The second dehydration involves three sequential, elementary steps: (a) hydride transfer from C1 to C2, (30 ?4); (b) proton transfer from O1 to O3 to form an oxonium ion, (4?5); and (c) dehydration, (5?6). The proton transfer from O1 to O3 is not feasible without prior hydride transfer (step (a)). Scanning along the nascent O3–H bond produces a potential of mean force with a very high barrier of over 667 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 H H HO OH O H HO HO H+ OH OH H H +OH 2 H HO O - H2O H HO H H HO H O+ H OH OH HO +OH O H OH H O O + HO O H O H H H 2 -H2O H H HO+ O O+ H HO H H 6 +OH 5 HO H O HO H HO O H 4 HO OH OH 3 H 3' C+ H H 2 HO H HO H HO OH H H 1 H HO OH O H H H 8 7 HO HO O O H2+O H H H HO O O+ - H2O O O - H+ H H H 8' 9 H H 10 Figure 2. Reaction mechanism for the dehydration of fructose to HMF. Unstable species are shown within square brackets. 80 kB T (58 kcal/mol) (shown in Fig. S1 of the Supplementary data scan failed to identify the formation of a stable intermediate, which is reflected in the lack of a local minimum in the pmf at short distances. The necessary hydride transfer (30 ?4) is not a fast one. The corresponding potential of mean force, shown in Figure 4a, has DF z ¼ 35 kBT (25 kcal/mol)—a significant activation energy to make the transfer a slow process but not high enough to render it unlikely, especially at elevated temperatures. The hydride transfer is electronically favoured, as we find that in the 30 ?4 transition the unbiased expectation of the quantum mechanical energy of the reacting QM ¼ 16:1 kB T (or 11.6 kcal/mol). system drops, with DE Considering, however, the free energy of the system, we see that this step requires DF ¼ þ19:6 kB T (14.1 kcal/mol). In the intermediate that forms, the C1–O1 bond length is shortened and fluctuates about the typical double bond length of 1.25 Å, while the C2–O5 double bond is elongated and fluctuates about the mean value of 1.4 Å, a consequence of the C2 carbon’s acquiring more of an sp3 character. Upon transferring to the C2 carbon, the hydride may be syn- or anti-coplanar to the OH group of the C3 carbon (or equivalently, the C2–C1 and C3–O3 bonds may be anti- or syn-coplanar). Structure optimization at the PM3 level, in the presence of explicit solvent, shows that the two conformations are isoenergetic. Furthermore, we have found that there are no differences between the respective free-energy profiles for the transfer of the hydride to ‘syn’ or ‘anti’ geometry. In Figure 4b we show free-energy profiles for the subsequent proton transfer (4?5) in four cases: (i) in the conformation where the C3–O3 and C2–C1 bonds anti-coplanar; (ii) in the syn-coplanar conformation; (iii) in the syn-coplanar conformation with a bridging water molecule in the QM region (to facilitate proton transfer); and (iv) in the syn-coplanar arrangement with two bridging water molecules in the QM region. They are all graphed as functions of the O3–H separation, namely, the length of the nascent bond. In 668 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 20 20 C2--OH C3--OH C4--OH 15 C2--O2 C3--O3 C4--O4 10 ΔF/kBT ΔF/kBT 0 10 5 -10 -20 0 -30 1 1.2 1.4 1.6 1.8 2 2.2 rxn coordinate (Å) -40 1 2 3 4 5 6 rxn coordinate (Å) Figure 3. Protonation and first dehydration free energies. (a) Free-energy profiles for the protonation of the ring OH groups of D-fructose. The reaction coordinate is the separation between the transferring proton and the acceptor oxygen atom; the reaction proceeds from right to left. (b) Free-energy profiles for the first dehydration, for each of the three OH groups on the ring of D-fructose. The reaction coordinate is the cleaving bond. (Free energies reported in thermal units, kBT, at T = 363 K.) cases (iii) and (iv), the extra water molecules were treated quantum mechanically. Without bridging water, the rates of proton transfer are comparable in the ‘syn’ and ‘anti’ conformations. In the syn-coplanar geometry (curve (ii) in Fig. 4b), the energy barrier is 42.2 kBT (30.5 kcal/mol). In the anti-coplanar arrangement, the barrier is equal to 41.5 kBT (29.9 kcal/mol) (curve (i) in Fig. 4b). Both curves have a local minimum (deeper in the case of the ‘syn’ geometry) at 1.8 Å, which corresponds to the formation of an intramolecular hydrogen bond between the acceptor OH on C3 and the donating one on C1. With respect to the reaction free energies, we find DF anti ¼ 23:8 kB T and DF syn ¼ 30:2 kB T. It is not uncommon for proton-transfer reactions of this type to be facilitated by mediating water molecules. To investigate such a possibility, we have included a water molecule in the QM region and allowed it to move freely (viz., unrestrained) within a domain defined by the intersection of two spheres: one centred about the O1 oxygen and a second centred about the O3 oxygen. The spheres have the same radius of 3.2 Å and soft, harmonic boundaries 2 jðr r0 Þ2 =2, with r0 ¼ 3:2 Å and j ¼ 100 kcal=ðmol Å Þ. (The radius of the spheres was chosen on the basis of the first coordination shell of hydrogen-bonded water molecules.) The relevant free-energy profile for proton transfer in the ‘syn’ geometry is curve (iii) in Figure 4b. Clearly, the water-assisted proton transfer results in an activation free-energy drop 18.4 kBT (13.3 kcal/mol), bringing the barrier down to DF z ¼ 23:8 kB T (17.2 kcal/mol). It also makes the transfer thermodynamically more favourable, with DF ¼ 12 kB T (8.7 kcal/mol). Note that the water molecule also stabilizes the state at 1.8 Å in which the transferring proton is shared between the donating O1 oxygen, the accepting O3 oxygen and the bridging water molecule via hydrogen bonds. We repeated this calculation in the presence of two extra water molecules in the QM region. Remarkably, we find that the activation free energy climbs back up to higher values, as shown by curve (iv) in Figure 4b, with DF z ¼ 36:4 kB T (26.3 kcal/mol). So, vicinal water accelerates the proton transfer by participating in the reaction, but more of it has the same adverse effect as not having any at all. This behaviour appears to be in agreement with Kuster’s observations that water displacement (by polyethylene glycol or other co-solvents) can accelerate the formation of HMF from D-fructose.6 In Figure 4c we show the potential of mean force for the second dehydration (5?6) as a function of the C3–O3 bond length. The activation free energy is equal to 34.3 kBT (24.7 kcal/mol). The local minimum at 2.5 Å corresponds to the formation of a hydrogen bond between the departing water molecule and the OH group on the C4 carbon. Removal of the second water requires DF ¼ 24:4 kB T (17.6 kcal/mol). We should note that for this step z 70 kB T, indicating how solvation, and in parwe calculate DE QM ticular entropic effects, reduce the size of the barrier. The species 6 is unstable, and once the second water molecule has broken off the ring, a spontaneous hydride transfer from the C4 to the C3 carbon (6?7) occurs, releasing 50.0 kBT (36.1 kcal/ mol) of free energy. (This was calculated by carrying out free-energy calculations for the reverse reaction 7?6, viz, hydride transfer from the C3 to the C4 carbon.) Thus, the C3 carbon retains its sp3 character, whereas the C4 acquires sp2 hybridization. To study this spontaneous hydride transfer further, we ran a series of QM/MM MD trajectories with the H atom constrained to stay on the C4 carbon after the water removal from C3. Within a few hundred femptoseconds—and following a charge transfer from the C1 to the C3 carbon—the HCO moiety (formyl group) broke off the C2 carbon of the ring. The reason for this very rapid H- transfer is that the second dehydration creates a partial electron deficit on the C3 carbon. The H- transfer from the C4 carbon to C3 replenishes this deficit, while, in turn, the electron deficit on C4 is covered by the non-bonding electrons of O4. If, instead, we constrain the H atom to remain on C4, then electronic density transfers from C2 to C3, weakening the C2–C1 bond and the HCO break-off ensues. We thus infer that the second dehydration is followed by two, very fast and probably competing processes. One involves hydride transfer from a neighbouring carbon atom to the one that has just lost an OH group. The other involves re-distribution of the electronic density, entailing partial fragmentation and thus decreased selectivity in the conversion of D-fructose to HMF. 3.1.3. Third dehydration and proton release The third dehydration involves the following elementary steps: (a) hydride transfer to C4 to restore its sp3 hybridization, (7?8); (b) proton transfer from C3 to O4 to form an oxonium ion, (8?80 ); and (c) the actual water break-off from C4, (80 ?9). The proton transfer to the O4 oxygen cannot take place without prior hydride transfer to C4—an already familiar situation from the second dehydration step. The relevant free-energy profile is shown in Figure S2 in the Supplementary data. A hydride transfer from C3 to C4 is not a consideration, on account of what we have just 669 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 40 50 (i) C2-C1 and C3-O3 bonds anti-coplanar (ii) C2-C1 and C3-O3 bonds syn-coplanar (iii) "syn" geometry w/ H2O bridge - H transfer from C1 to C2 40 30 (iv) "syn" geometry w/ two H2O in bridge ΔF/kT ΔF/kT 30 20 (ii) (iv) 20 (i) 10 10 (iii) 0 0 1 1.5 2 0 2.5 1 2 3 4 5 rxn coordinate (Å) rxn coordinate (Å) 10 ΔF/kT 0 -10 -20 1 1.5 2 2.5 3 3.5 4 4.5 rxn coordinate (Å) Figure 4. Second dehydration free energies. (a) Potential of mean force for hydride transfer from the C1 carbon atom to the C2. The reaction is monitored by the separation between the transferring hydride and the acceptor carbon atom C3 and proceeds from right to left. (b) Potential of mean force for proton transfer from the O1 oxygen atom to the O3. The reaction coordinate is the separation between the transferring proton and the acceptor oxygen atom (O3) and the direction of reaction is from right to left. We graph the free-energy profiles for four cases: (i) C3–O3 and C2–C1 bonds are anti-coplanar; (ii) C3–O3 and C2–C1 bonds are syn-coplanar; (iii) syn-coplanar conformation with one mediating water molecule in the QM region; and (iv) syn-coplanar geometry with two mediating water molecules in the QM region. (c) Potential of mean force for the second dehydration. The reaction coordinate is the cleaving C3–O3 bond. (Free energies reported in thermal units, kBT, at T = 363 K.) discussed regarding the second dehydration. Thereby, there are only two alternatives available to us: hydride transfer from C5 to C4; or simultaneous H- transfer from C3 to C4 and from C2 to C3, which would force the ring oxygen, O5, to partake in electron sharing with its non-bonding electrons and which would result in an electronic configuration that does not weaken the C2–C1 bond. The coordinated H- transfer from C3 to C4 and from C2 to C3 does indeed result in a stable intermediate without fragmentation products. However, the activation free energy for this elementary reaction is 68 kBT (49 kcal/mol), which precludes the possibility of measuring any conversion on a laboratory time scale, even at elevated temperatures; the relevant free-energy profile is shown in Figure S3 in the Supplementary data. Nevertheless, hydride transfer from the C5 carbon to the C4 (7?8) is a viable alternative with DF z ¼ 44:0 kB T (31.8 kcal/mol) (see Fig. 5a). It results in a stable intermediate with the C5 carbon being double-bonded to the O5 oxygen. This is a significant activation energy at T = 363 K, but not so much for temperatures in the range between 393 and 450 K, which are typical reaction temperatures for the conversion of D-fructose to HMF. This step requires DF ¼ 30:3 kB T (21.9 kcal/mol). Subsequent capturing of a C3-hydrogen by the OH group on the C4 carbon, 8?80 , requires 32.0 kBT (23.1 kcal/mol) of activation energy. The relevant potential of mean force is graphed in Figure 5b as a function of the distance between the transferring hydrogen and the acceptor oxygen. As expected, the H capture by the O4 atom is endothermic with DF ¼ 12:8 kB T (9.2 kcal/mol), as the C–H bond is stronger than the O–H bond. We have investigated whether a bridging water molecule can facilitate the H transfer from C3 to O4 and found that the free-energy barrier for such a process is more than twice as high [75.0 kBT (54.1 kcal/mol)]. The resulting intermediate, 80 , is unstable. As soon as the proton is captured by the O4 oxygen, the C4–O4 bond cleaves and the incipient water molecule spontaneously departs (80 ?9). Thus, the third dehydration is non-activated and quite exothermic, releasing 70:0 kB T (50.5 kcal/mol) of free energy (see Fig. 5c). 670 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 40 30 ΔF/kT ΔF/kT 30 20 10 20 10 0 0 1 2 1.5 1 2.5 2 1.5 rxn coordinate (Å) 2.5 3 rxn coordinate (Å) 20 70 60 15 40 ΔF/kT ΔF/kT 50 30 20 10 5 10 0 1 1.5 2 2.5 3 3.5 4 4.5 rxn coordinate (Å) 5 0 1 1.5 2 2.5 3 rxn coordinate (Å) Figure 5. Third dehydration and proton release free energies. (a) Free-energy profile for H-transfer from the C5 carbon to C4. The reaction coordinate is the distance between the transferring H atom and the acceptor carbon (C4) and the direction of reaction is from right to left. (b) Free energy profile for H transfer from the C3 carbon to the O4 oxygen atom and water formation. The reaction coordinate is the separation between the transferring H atom and the acceptor oxygen O4. (c) Free-energy profile for the third dehydration. The reaction coordinate is the breaking bond. The process is non-activated and very exothermic. (d) Free-energy profile for proton transfer from the HMF molecule to a solvent water molecule. (Free energies reported in thermal units, kB T, at T = 363 K.) Proton (i.e., catalyst) release from HMF back to the solvent is an activated process, with DF z ¼ 15 kB T (10.8 kcal/mol) and DF ¼ 4:3 kB T (3.1 kcal/mol). The relevant pmf is shown in Figure 5d. 3.2. Effects of solvent dynamics We have already noted instances where solvation dynamics affect the energetics of the reaction. Here, we present a more detailed analysis on select elementary steps where these effects are more pronounced. In Figure 6 we have analysed the pmf for the first water removal into energy and entropy contributions, DF ¼ DU T DS; and the QM þ DU vdW þ DU q . thermodynamic energy further as DU ¼ DE QM is the unbiased conditional expectation of the energy (PM3) DE of the system in the QM region—the part of the energy that is calculated quantum mechanically and, in analogy, the energy one would obtain if there was no solvent in the system and the reaction was ‘run’ in vacuo. DU vdW is the van der Waals component of the total energy of the system and includes interactions between the MM solvent molecules and interactions between the solvent molecules and the molecules in the QM region. Similarly, DU q denotes the electrostatic interactions. QM alone, we Looking at the first dehydration by considering DE see that it has a high barrier to reaction and is surely endoergic. The significantly low barrier in DF is primarily due to solvent reorganization. What drives this re-organization is the need of the system to optimize unfavourably strong Coulomb interactions that are generated by the charge re-distribution that occurs in fructose as the C2–O2 is breaking and a water molecule departs. The solvent re-organization does not appear to be entropically favoured, as we see the term T DS to rise. It also leads to some solvent ‘packing’ around the fructose molecule, which one infers from the concomitant increase in DU vdW . Thus, the barrier in DF could be characterized as an entropic one. As fragmentation in the gas phase is entropically favourable, the observed entropic barrier must be attributed to the presence of the solvent. The elementary reactions that involve intramolecular hydride transfer are among the slowest in the mechanism. An analysis of the free-energy profile for H- transfer from C1 to C2 (3?4 in Fig. 2) is shown in Figure 7. Here, the reaction proceeds from large 671 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 ΔF ΔU -T ΔS ΔEQM Mulliken partial charges 50 C1 0.4 Energy (kBT units) ΔUvdW ΔUq 0 C2 0.2 hydride 0 ring-O -0.2 O1 -50 -0.4 1 2 1.5 Hydride -- Accepting carbon separation (Å) 1 2 3 4 5 C2--O2 separation (Å) Figure 6. First dehydration energy profiles. The potential of mean force, DF, is analysed into thermodynamic energy, DU, and entropy, T DS, contributions. The QM , and energy DU is further analysed into the quantum mechanical contribution, DE the van der Waals, DU vdW , and electrostatic, DU q , interaction energies. values of the reaction coordinate (H–C2 separation) to small ones and we clearly see that the free-energy barrier is mostly due to DU—specifically, the electrostatic component DU q . In fact, the situation appears typical of charge transfer reactions between two redox states with asymmetrical charge distributions (see Fig. 8). In a hydride transfer, the H atom is carrying an extra electron, which must transfer from one carbon to another or from one redox state to another. For that to occur we must first have symmetrization of the wave function at the transition state or, in other words, the two localized states (orbitals) situated on the two carbon atoms must become degenerate, neither preferred energetically. Only then will the quantum mechanical probability for electron transfer be large. From the perspective of the solvent, the polar environment organizes around the asymmetrical charge distribution in such a way as to give favourable solvation energy. The loss of asymmetry in the charge distribution will, therefore, be associated with a loss of that favourable energy. This means that passing from a state in which electronic charge is localized on one centre to a state where the charge is distributed over two centres (symmetrization) will be associated with an increase in energy. The latter, resonating state, Figure 8. Charge asymmetry between the two stable states in the hydride transfer described by the step 3?4 of the mechanism. is the transition state for hydride transfer, and the increase in energy to reach it is the activation energy for the hydride transfer. Once attained through some random thermal fluctuation of the environment, the system—reacting fructose plus environment— can move downhill in energy or free energy either by falling back to where it started, or by falling towards the other charge localized state. In the latter case, hydride transfer occurs. So, the hydride transfer occurs due to environmental fluctuations. The energy required for the transfer must come from the bath—the surrounding solvent—and it entails a fluctuation of the molecular environment. It is exactly this temporal re-organization that creates a situation where the two redox states are degenerate. At such a stage, the H will resonate between both carbon sites and it will localize again only after the environment moves towards one of the two solvated configurations. The situation is exactly the same in the case of the hydride transfer in the step 7?8 of the mechanism, which is rate-limiting. 4. Conclusion We have studied the energetics of the closed-ring mechanism for the acid-catalysed dehydration of D-fructose to HMF in an aqueous medium by calculating free-energy profiles for all the elementary steps of the reaction. A schematic of the overall energy profile is given in Figure 9. We have found that the protonation of each of 80 50.0 ΔF ΔU -T ΔS ΔEQM 30.0 ΔF (kcal/mol) Energy (kBT units) 60 ΔUvdW 40 24.8 40.0 ΔUq 20 (6) 21.2 (8’) 31.8 17.2 (8) 20.0 10.0 0.0 (5) 25.3 -50 -70 (4) 12.8 FRU 6 (2) (7) (3, 3’) -10.0 10.8 -20.0 (9) 0 -20 1 1.2 1.4 1.6 1.8 2 2.2 -30.0 HMF rxn coordinate Hydride -- accepting carbon separation (Å) Figure 7. Energy profiles for hydride transfer (3?4) from C1 to C2, after the first and before the second dehydration. Figure 9. Overall free-energy profile for the dehydration of D-fructose to HMF, in units of kcal/mol. Energy differences are marked in red. Species labelling as in Figure 2. 672 S. Caratzoulas, D. G. Vlachos / Carbohydrate Research 346 (2011) 664–672 the three OH groups of the ring, which initiates the reaction, is equally likely. However, subsequent removal of the protonated OH of the C2 carbon is much faster and thermodynamically more favourable than the removal of the other two during the first dehydration of fructose. If the dehydration is initiated at the OH group on the C4 carbon, subsequent steps lead either to ring opening or fragmentation products. For the pathway initiated at the OH on C2, the second dehydration may be followed either by a hydride transfer to stabilize the resulting intermediate or a charge transfer, which leads to fragmentation and release of a formyl group (HCO). There is a considerable energetic cost to the removal of the first two water molecules, and the rate of re-hydration is significant, as can been seen in Figure 9. However, the reaction becomes less reversible after the second dehydration is complete. We have found that the proton transfer in 4?5 is substantially accelerated if it is water-mediated. However, we have also found that too much water in the vicinity of the transferring proton has an adverse effect. On the other hand, the proton transfer in 8?80 is not affected by the presence of water. From a micro-kinetic point of view, the slower steps of the reaction are those that involve intramolecular hydride transfer. We have shown that the high barriers involved in hydride transfer are due to re-organization of the polar solvent environment and the solvation of asymmetrical electronic charge distributions. The observed solvent effects are in agreement with Kuster’s reporting that water displacement (by polyethylene glycol or other co-solvents) can accelerate the formation of HMF from D-fructose.6 The rate-limiting step is the hydride transfer from the C5 to the C4 carbon prior to the third dehydration, requiring 31.8 kcal/mol of activation free energy. This amounts to a CLTST thermal rate k ¼ ðkB T=hÞ expðDF z =kB TÞ 5:8 107 s1 . Kinetics studies by Moreau et al. report apparent activation energies of 33.7 kcal/mol in zeolites.25 Using their value for the Arrhenius pre-exponential factor, the respective conversion rate constant may be estimated at 1.5 107 s1, at 363 K. The order of magnitude agreement is very encouraging. Finally, according to our calculations, for the overall reaction DF ¼ 20:5 kcal=mol, at 363 K. We are not aware of any experimental values of the free energy of conversion of fructose to HMF. Thermochemistry studies by Assary et al. at the G4 theory level and implicit solvation report DG ¼ 32:4 kcal=mol at 298 K.26 Acknowledgements We wish to thank Professor Doug Doren for useful discussions. This work was supported as part of the Catalysis Center for Energy Innovation, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Basic Energy Sciences under Award Number DE-SC0001004. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.carres.2011.01.029. References 1. Röper, H. Starch/Stärke 2002, 54, 89–99. 2. Chheda, J.; Huber, G.; Dumesic, J. Angew. Chem., Int. Ed. 2007, 46, 7164–7183. 3. Werpy, T.; Petersen, G. Top Value Added Chemicals from Biomass, Vol. 1: Results of Screening for Potential Candidates from Sugars and Synthesis Gas; 2004. Available electronically at http://www.osti.gov/bridge. 4. Huber, G.; Chheda, J.; Barrett, C.; Dumesic, J. Science 2005, 308, 1446–1450. 5. Kuster, B.; van der Baan, H. Carbohydr. Res. 1977, 54, 165–176. 6. Kuster, B. Carbohydr. Res. 1977, 54, 177–183. 7. Kuster, B.; Temmink, H. Carbohydr. Res. 1977, 54, 185–191. 8. Kuster, B. Starch/Stärke 1990, 42, 314–321. 9. Amarasekara, A.; Williams, L.; Ebede, C. C. Carbohydr. Res. 2008, 343, 3021– 3024. 10. Antal, M. A., Jr.; Mok, W. Carbohydr. Res. 1990, 199, 91–109. 11. Bicker, M.; Kaiser, D.; Ott, L.; Vogel, H. J. Supercrit. Fluids 2005, 36, 118–126. 12. Roman-Leshkov, Y.; Chheda, J.; Dumesic, J. Science 2006, 312, 1933–1937. 13. Moliner, M.; Román-Leshkov, Y.; Davis, M. Proc. Natl. Acad. Sci. U.S.A. 2010. doi:10.1073/pnas.1002358107. 14. Case, D. A.; Darden, T. A.; Cheatham, T. E.; III, Simmerling, C.L.; Wang, J.; Duke, R. E.; Luo, R.; Merz, K. M.; Pearlman, D. A.; Crowley, M.; Walker, R. C.; Zhang, W.; Wang, B.; Hayik, S.; Roitberg, A.; Seabra, G.; Wong, K. F.; Paesani, F.; Wu, X.; Brozell, S.; Tsui, V.; Gohlke, H.; Yang, L.; Tan, C.; Mongan, J.; Hornak, V.; Cui, G.; Beroza, P.; Mathews, D. H.; Schafmeister, C.; Ross, W. S.; Kollman, P. A. AMBER 9, University of California, San Francisco; 2006. 15. Wu, R.; Xie, H.; Cao, Z.; Mo, Y. J. Am. Chem. Soc. 2008, 130, 7022–7031. 16. Pomata, M.; Sonoda, M.; Skaf, M.; Elola, M. J. Phys. Chem. B 2009, 113, 12999– 13006. 17. Ferrenberg, A.; Swendsen, R. Phys. Rev. Lett. 1989, 63, 1195–1198. 18. Kumar, S.; Bouzida, D.; Swendsen, R.; Kollman, P.; Rosenberg, J. J. Comput. Chem. 1992, 13, 1011–1021. 19. Mok, W. S.; Antal, M. J., Jr.; Varhegyi, G. Ind. Eng. Chem. Res. 1992, 31, 94–100. 20. Nimlos, M.; Qian, X.; Davis, M.; Himmel, M.; Johnson, D. J. Phys. Chem. A 2006, 110, 11824–11838. 21. Qian, X.; Nimlos, M.; Davis, M.; Johnson, D.; Himmel, M. Carbohydr. Res. 2005, 340, 2319–2327. 22. Kramers, H. A. Physica 1940, 4, 284–304. 23. Hänggi, P.; Talkner, P.; Borkovec, M. Rev. Mod. Phys. 1990, 62, 251–341. 24. Benderskii, V.; Makarov, D.; Wight, C. Adv. Chem. Phys. 1994, 88, 1–360. 25. Moreau, C.; Finiels, A.; Vanoye, L. J. Mol. Catal. A: Chem. 2006, 253, 165–169. 26. Assary, R.; Redfern, P.; Hammond, J.; Greeley, J.; Curtiss, L. J. Phys. Chem. B 2010, 114, 9002–9009.
© Copyright 2026 Paperzz