Mechanism of Br?nsted acid-catalyzed glucose dehydration

Article abstract of DOI:10.1002/cssc.201403264

We present the first DFT-based microkinetic model for the Br?nsted acid-catalyzed conversion of glucose to 5-hydroxylmethylfurfural (HMF), levulinic acid (LA), and formic acid (FA) and perform kinetic and isotopic tracing NMR spectroscopy mainly at low conversions. We reveal that glucose dehydrates through a cyclic path. Our modeling results are in excellent agreement with kinetic data and indicate that the rate-limiting step is the first dehydration of protonated glucose and that the majority of glucose is consumed through the HMF intermediate. We introduce a combination of 1) automatic mechanism generation with isotopic tracing experiments and 2) elementary reaction flux analysis of important paths with NMR spectroscopy and kinetic experiments to assess mechanisms. We find that the excess formic acid, which appears at high temperatures and glucose conversions, originates from retro-aldol chemistry that involves the C6 carbon atom of glucose.

Full text of DOI:10.1002/cssc.201403264

DOI: 10.1002/cssc.201403264
Full Papers
Mechanism of Brønsted Acid-Catalyzed Glucose
Dehydration
Liu Yang, George Tsilomelekis, Stavros Caratzoulas, and Dionisios G. Vlachos*[a]
We present the first DFT-based microkinetic model for the
Brønsted acid-catalyzed conversion of glucose to 5-hydroxyl-
methylfurfural (HMF), levulinic acid (LA), and formic acid (FA)
and perform kinetic and isotopic tracing NMR spectroscopy
mainly at low conversions. We reveal that glucose dehydrates
through a cyclic path. Our modeling results are in excellent
agreement with kinetic data and indicate that the rate-limiting
step is the first dehydration of protonated glucose and that
the majority of glucose is consumed through the HMF inter-
mediate. We introduce a combination of 1) automatic mecha-
nism generation with isotopic tracing experiments and 2) ele-
mentary reaction flux analysis of important paths with NMR
spectroscopy and kinetic experiments to assess mechanisms.
We find that the excess formic acid, which appears at high
temperatures and glucose conversions, originates from retro-
aldol chemistry that involves the C6 carbon atom of glucose.
Introduction
Recent interest in renewable fuels and chemicals has drawn at-
tention to the conversion of lignocellulosic biomass to plat-
form molecules,[1] such as 5-hydroxymethyl furfural (HMF).
Most research has focused on the conversion of fructose to
HMF,[2] a very versatile platform toward fuels and chemicals.
Fructose dehydration is fairly easy to perform and is catalyzed
by homogeneous and heterogeneous Brønsted acid catalysts.[3]
In contrast, it is hard to convert glucose to HMF in acidic,
aqueous media; the selectivity to HMF[4] is low unless a catalyst
is used to first convert glucose to fructose.[5] Although the
mechanism of fructose dehydration has received considerable
attention and consensus is forming that it does not involve
ring opening,[2,6] the reaction mechanism of glucose dehydra-
tion is still elusive.[7] Although HMF formation may proceed
through an acyclic mechanism,[7b,8] cyclic pathways have also
been suggested by theoretical studies.[7b–d] The cyclic mecha-
nism involves the direct transformation of the pyranose ring to
a five-membered ring intermediate.[8a,9] The alternative acyclic
mechanism proposes glucose isomerization to fructose and
subsequent dehydration to HMF.[4,7f,8c–e,10] In acidic, aqueous
media, the HMF molecule is rehydrated easily to form levulinic
acid (LA) and formic acid (FA), and LA is considered as another
cohol (FAL) with the concomitant release of formic acid. The
thermochemistry of this pathway has been investigated com-
putationally,[7b] but
a detailed mechanism has not been
mapped out, and kinetic data, that is, activation energies,
which would help us assess its relative contribution and
whether it is kinetically feasible, are not available.
In this article, we elucidate the kinetics of the Brønsted acid-
catalyzed dehydration of glucose by setting up the most com-
prehensive, to date, reaction network that allows multiple
pathways to be evaluated using microkinetic modeling and re-
action path analysis. The reaction network considers the cyclic
and acyclic (viz., through fructose) conversion of glucose to
HMF, glucose condensation, the conversion of glucose to FAL
and FA, and the degradation of HMF to LA and FA and of FAL
to LA. Plausible pathways were also generated using automatic
mechanism generation and path identification algorithms, and
DFT calculations were performed to locate transition states and
to calculate the free energies of reaction and activation. The
predictions of the microkinetic model were tested and comple-
mented with isotopic tracing studies using NMR spectroscopy.
important, cellulose-derived platform chemical. Interestingly, Results and Discussion
the dehydration of glucose, under the same conditions, yields
Reaction mechanisms
FA and LA at a ratio higher than 1:1. This implies that glucose
dehydration does not proceed solely to HMF and that another
pathway must also exist that yields FA in the products. One
such pathway entails the dehydration of glucose to furfuryl al-
In the following we present DFT calculations for the conversion
of glucose to FAL and to the condensation product 1,6-O-a-d-
glucopyranosyl-d-glucose.
[a] Dr. L. Yang, Dr. G. Tsilomelekis, Dr. S. Caratzoulas, Dr. D. G. Vlachos
Department of Chemical and Biomolecular Engineering
Catalysis Center for Energy Innovation, University of Delaware
Newark, DE 19716 (USA)
First dehydration of protonated glucose
Brønsted acid-catalyzed glucose dehydration is initiated by the
protonation of hydroxyl groups. Although the different hydrox-
yl groups show similar protonation energetics in aqueous me-
dium,[7b] the direct dehydration of glucose to FAL is initiated
E-mail: vlachos@udel.edu
Supporting Information for this article is available on the WWW under
http://dx.doi.org/10.1002/cssc.201403264.
ChemSusChem 0000, 00, 0 – 0
1
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
Scheme 1. Labeling of glucose and fructose atoms.
Figure 2. Free energy profile at 298.15 K for glucose dehydration to FAL
through intermediate A in acid solution. All energies are referred to free glu-
cose in water. Transition states are shown in Figure S2.
by the protonation of C2ꢀOH (see Scheme 1 for labeling).
Recent theoretical investigations have suggested that the re-
moval of a water molecule from the protonated glucose results
in the rearrangement of the six-membered ring to a five-mem-
bered ring.[7b–d] Upon dehydration at the C2 position, either
the C3ꢀC4 or the C1ꢀO5 bond can cleave to yield two rear-
rangement products (hereafter referred to as intermediates A
and B). These two intermediates are nearly kinetically and ther-
modynamically equivalent (Figure 1). Notably, according to
Scheme 2. Intermediates that form during glucose dehydration to FAL
through intermediate A in acid solution. The glucose C3 carbon atom that
ends up in FA is marked with a red star.
free energy is 5.5 kcalmolꢀ1. The proton on the gem-diol inter-
mediate may then migrate to the adjacent C3ꢀOH group,
which is slightly thermodynamically uphill (from S3 to S4 in
Figure 2). Intermediates S3 and S4 adapt a gem-diol structure;
A similar structure has been reported before for the rehydra-
tion reaction of FAL in acidic solution.[11]
The dehydration of S4 produces an unstable carbocation at
the C3 position, which undergoes spontaneous deprotonation
to a neighboring water molecule. The overall activation and re-
action free energies for the dehydration of S4 and the subse-
quent deprotonation are 18.4 and ꢀ14.1 kcalmolꢀ1, respective-
ly. The thermodynamically stable intermediate S5 can be pro-
tonated easily at the C1ꢀOH position, which requires a mere
3.0 kcalmolꢀ1 of activation free energy, and then undergo de-
hydration. Finally, FAL is generated through CꢀC scission, with
the concomitant release of a molecule of FA. Thus, we see that
the highest point along the free energy profile is the transition
state for the first dehydration, which has an apparent activa-
tion free energy of 31.1 kcalmolꢀ1 (Figure 2). This value is con-
sistent with the glucose dehydration barrier of 30–35 kcal
molꢀ1 reported previously and is potentially the rate-determin-
ing step.[12] We return to this point in the next section, in
which we develop a microkinetic model for the entire reaction
network and analyze its predictions. This reaction pathway pro-
duces FAL, which subsequently undergoes reactions to pro-
duce LA and other byproducts. From the reaction stoichiome-
try, the ratio of FA and FAL is 1:1. As the other byproducts pro-
duced from FAL limit the yield of LA, this may result in
a higher productivity of FA than LA.
Figure 1. Free energy profile for the removal of a water molecule from glu-
cose protonated at the C2ꢀOH position and the formation of two different
intermediates A and B.
recent calculations, intermediate B can also form during the
dehydration of fructose to HMF.[2a] This indicates that interme-
diate B is a bifurcation point at which glucose dehydration can
follow either the pathway to HMF or the pathway to FAL. In
the following we will ascertain whether kinetics or thermody-
namics determine which pathway will be followed.
Reaction pathway through intermediate A
The free energy profile for glucose conversion to FAL through
intermediate A is shown in Figure 2. This path does not lead to
HMF formation. The overall process is exothermic by 46.4 kcal
molꢀ1. The protonated carbonyl group in intermediate A (S2 in
Figure 2 and Scheme 2) can be hydrated easily to form the
gem-diol structure S3 (Scheme 2). For this hydration reaction,
the activation free energy is 8.3 kcalmolꢀ1 and the reaction
&
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
2
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
Assary et al.[7b] proposed a possible pathway for FAL forma-
tion based on thermochemical calculations alone. The reaction
mechanism and free energy profile presented in this article
differ from those of Assary et al., and our DFT calculations sug-
gest different reaction intermediates and energetics. Our reac-
tion system includes two explicit water molecules, the pres-
ence of which reveals a different reaction pathway.
previously for fructose dehydration to HMF through interme-
diate B is shown in Scheme 3.[2a] Given that intermediates A
and B are structurally similar, we have investigated whether in-
termediate B can also be involved in the conversion of glucose
to FAL. This reaction pathway is presented in Figure 3, and the
associated intermediates and transition states are presented in
Scheme 4.
+
According to this mechanism, breaking the C2ꢀ OH2 bond
and the formation of intermediate B requires 25.2 kcalmolꢀ1 of
activation free energy (TS10 and S11 in Scheme 4). Similar to
the reaction pathway that proceeds through intermediate A,
the hydration of intermediate B yields the gem-diol intermedi-
ate S12 (Figure 3 and Scheme 4). Subsequent proton migration
to C3ꢀOH results in the intermediate S13, the dehydration of
which also leads to the release of a FA molecule, in a concerted
fashion (TS12 in Scheme 4). This process requires an activation
free energy of 25.6 kcalmolꢀ1 and releases 32.6 kcalmolꢀ1 of re-
action energy. This reaction is highly exothermic because of
the formation of the thermodynamically stable intermedi-
ate S14. Finally, the formation of FAL requires the protonation
and dehydration of S14 with activation energies of 2.5 and
4.4 kcalmolꢀ1, respectively. An inspection of the overall energy
profile reveals that the apparent activation free energy to FAL
through intermediate B is 31 kcalmolꢀ1, which is similar to that
of going through intermediate A. Thus, this is another poten-
tially important reaction pathway in glucose dehydration.
In acid solution, FAL is very reactive and forms LA through
rehydration. As FAL can also form polymers and other byprod-
ucts, the yield of LA is limited. If we consider the stoichiometry
of glucose dehydration to FAL (glucose!FAL+FA+2H2O), one
would expect that the FAL molar yield would be higher than
that of LA. Also, if FA and LA were formed solely through HMF
rehydration, then the FA/LA ratio should be 1:1. In acidic
media, FA is detected at a higher than 1:1 ratio, and the pro-
posed pathway to FAL can explain the excess FA formation.
As we have indicated, the intermediate B is a bifurcation
point at which glucose dehydration can follow either the path-
way to HMF or the pathway to FAL. The reaction pathway
from intermediate B to HMF has been investigated by Carat-
zoulas and Vlachos,[2a] as it is part of fructose dehydration to
HMF (Scheme 3). A comparison between these two reaction
channels, after the formation of intermediate B (Figure S4), in-
dicates that both pathways are plausible and that the pathway
to HMF is thermodynamically more favorable than that to FAL.
We will show that this picture is consistent with the
Reaction pathway through intermediate B
Previous studies indicate that the removal of two water mole-
cules from intermediate B (S11 in Figure 3 and Scheme 4) re-
sults in the formation of HMF.[2a] A reaction pathway explored
Figure 3. Free Energy profile at 298.15 K for glucose dehydration to FAL
through intermediate B in acid solution. All energies are referred to free glu-
Scheme 3. Reaction pathway from fructose to intermediate B
outcome of the microkinetic model and of reaction
path analysis.
Glucose condensation to 1,6-O-a-d-glucopyranosyl-
d-glucose
Glucose condensation is initiated by C1ꢀOH protona-
tion and water removal. Although the dehydration of
the C2 hydroxyl group results in rearrangement of
the six-membered ring to a five-membered one, de-
Scheme 4. Intermediates that form during glucose dehydration to FAL through inter-
mediate B in acid solution. The glucose C1 carbon atom that ends up in FA is marked
with a red star.
hydration of the C1 hydroxyl group yields a carbocat-
ion that is stabilized by the oxonium ion resonance
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
3
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
cose in water. Transition states are shown in Figure S3.
These are not the final page numbers! ÞÞ

Full Papers
a reaction network that comprises: 1) glucose isomerization to
fructose through protonation at the O6 site, which subse-
quently dehydrates to HMF (see Supporting Information for
more details); 2) direct glucose dehydration to HMF; 3) direct
glucose dehydration to FAL; 4) HMF degradation to LA and FA;
5) FAL to LA; 6) glucose condensation to 1,6-O-a-d-glucopyra-
nosyl-d-glucose; and 7) side reactions from glucose, fructose,
HMF, and FAL.
Even though all five hydroxyl groups of glucose can be pro-
tonated in acidic solution, glucose condensation initiated by
the protonation of O1 is less significant because of the low
glucose concentrations and the high reaction temperatures
considered in this work. Indeed, the inclusion of the condensa-
tion pathway in the reaction network decreases the overall
glucose dehydration rate slightly, but the FA, LA, and HMF
yields are practically unaffected (Figure S6). Protonation of O3,
O4, or O6 contributes very little to the overall reaction accord-
ing to our NMR spectroscopy and isotope experiments. Thus,
these reaction pathways were not considered in our MKM reac-
tion network.
Figure 4. Free energy profile for glucose condensation to 1,6-O-a-d-gluco-
pyranosyl-d-glucose through C1ꢀOH protonation at 298.15 K. The transition
states are shown in Figure S4. All free energies are with respect to glucose.
The experimental kinetics of glucose dehydration was exam-
ined in water at 1408C with an HCl concentration of
0.1 molLꢀ1. The concentration profiles of glucose, HMF, FA, and
LA are reported in Figure 5. A comparison of our model with
literature data at higher temperatures and various (lower) pH
values is shown in Figure S7. Evidently, the glucose dehydra-
tion rate depends strongly on the temperature and pH of the
reaction solution. Increasing (decreasing) the reaction tempera-
ture (pH) increases the reaction rate, the yield of HMF, and the
total yield of LA. Compared with both our own and published
kinetic studies (Figure S7), the model captures the reaction
chemistry under a broad range of conditions without adjusta-
ble parameters. However, at low glucose conversion and after
2 h, the experimental FA and LA yields are the same at 1408C
(solid points in Figure 5), whereas the model predicts an FA/LA
ratio slightly greater than 1. As the excess FA predicted from
the microkinetic model is produced by the reaction pathway
through intermediate A, the disagreement between the calcu-
structure S18 (Figure 4). All five hydroxyl groups of a second
glucose molecule can react with the C1 carbocation to form
a disaccharide, thus multiple products are possible. The yield
of the condensation products depends on the reactivity of an
individual hydroxyl group toward the carbocation as well as
the stability of the glycosidic bond. It has been shown experi-
mentally that the 1,6-linked disaccharide is the most favorable
product.[12] This is likely because of the reduced steric hin-
drance between the bulky sugar rings if there is an additional
methylene group in the linkage. As the carbocation is essen-
tially flat, the hydroxyl group can add to either side and form
a and b addition products. a-Linked disaccharides are pro-
duced more than their b isomers.[12] This is because the linkage
is equatorial to one ring and axial to the other in the a isomers,
which reduces the steric hindrance (Figure S5). Therefore, our
calculations focused on the a isomer of the 1,6-linked disac-
charide (1,6-O-a-d-glucopyranosyl-d-glucose), and the reaction
energetics at 298.15 K are shown in Figure 4. It can be seen
from the free energy profile that the protonated glucose is ap-
proximately 6.1 kcalmolꢀ1 higher in free energy than that of
the reactant. The subsequent breakage of the C1ꢀO1 bond re-
quires 12.8 kcalmolꢀ1 of activation free energy. The C1 carbo-
cation attacks the C6ꢀOH bond of another glucose molecule,
which forms the 1,6-linked disaccharide. This step presents
a barrier of 16.1 kcalmolꢀ1, followed by spontaneous deproto-
nation of the addition product. The apparent activation free
energy is approximately 20 kcalmolꢀ1. This barrier is lower
than that for glucose dehydration through C2ꢀOH protona-
tion. However, the free energy change of condensation is
15.6 kcalmolꢀ1, which makes this pathway less significant at
high temperatures.
First-principles-based microkinetic modeling (MKM)
Figure 5. Concentration profiles for major species in acid-catalyzed glucose
dehydration in a batch reactor at 1408C with 0.1 molLꢀ1 HCl and an initial
glucose concentration of 0.28 molLꢀ1. Solid lines represent calculated
energy profiles and points represent experimental data.
In addition to the DFT calculations and to compare with kinetic
studies,[4,12–13] we have constructed a microkinetic model for
&
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
4
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
lation and the experiments implies that, at low temperatures
or low conversions, the path through the intermediate A is less
important. We will address this issue in the next section.
To identify the rate-determining step(s), we performed sensi-
tivity analysis. The rate constants of all the elementary reac-
tions were perturbed by 10%, and the concentration of glu-
cose was recorded. The normalized sensitivity coefficient [NSC;
Eq. (1)] is shown in Figure 6.
in which Ri is the measured response and Ai corresponds to
the model parameter that is perturbed[14] for the five key ele-
mentary steps. Both reactions 3 and 4 affect the glucose dehy-
dration significantly, and reaction 12 is also kinetically impor-
tant. According to our reaction network, reactions 3 and 4 cor-
respond to the first dehydration of protonated glucose that
leads to the formation of intermediates B and A, respectively.
Reaction 12 corresponds to the hydride transfer involved in
the glucose isomerization to fructose. The free energy profiles
shown in Figure 2 and 3 are consistent with our MKM simula-
tions, which show that the first dehydration of glucose is the
rate-determining step.[2]
dðln RjÞ Ai dRj
ð1Þ
NSCi;j
¼
¼
Rj dAi
dðln AiÞ
The reversibility of various
steps was assessed by defining
the partial equilibrium ratio f as
i
[Eq. (2)]:
ꢀi ¼ rf,i=ðrf,i þ rr,iÞ
ð2Þ
in which rf and rr indicate the
forward and reverse reaction
rates. A value of f =0.5 indi-
i
cates that a reaction is in partial
equilibrium, whereas a value of
1 or 0 implies an irreversible re-
action (forward or reverse, re-
spectively). The calculated partial
equilibrium ratio of the first de-
hydration reaction that converts
protonated glucose to interme-
diates A and B is approximately
1.0, which is consistent with it
being rate-limiting.
Figure 6. Normalized sensitivity coefficients (NSCi,j) for glucose dehydration at 1408C with an acid concentration
of 0.1 molLꢀ1 at 30 and 60 min. Only the largest sensitivity coefficients are shown. The first dehydration of glucose
is the most kinetically significant elementary step.
Elementary reaction flux analy-
sis and isotopic tracing experi-
ments
The analysis of reaction fluxes in
kinetic models, also known as re-
action path analysis, allows us to
understand the sequence of mo-
lecular events and the impor-
tance of the corresponding ele-
mentary reactions. Here we
couple flux analysis with isotopic
tracing experiments to further
investigate reaction mechanisms
in a quantitative manner. We il-
lustrate this approach next.
The major reaction pathway in
the Brønsted acid-catalyzed glu-
cose dehydration is illustrated in
Scheme 5. The reaction flux anal-
ysis indicates that the isomeriza-
tion of glucose to fructose ap-
pears to have a negligible contri-
Scheme 5. Reaction path analysis of major elementary processes in Brønsted acid-catalyzed glucose dehydration.
Numbers on the arrows are contributions from different parallel reactions at 1408C with 0.1 molLꢀ1 HCl.
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
5
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
bution to the overall rate of glucose consumption. In other
words, the cyclic pathway dominates over the acyclic one.
Consistent with the DFT-calculated free energy barriers, the
contributions from the parallel pathways shown in Scheme 5
indicate that the formation of intermediate B is preferred over
that of intermediate A at 1408C and that most of the FA and
LA is produced by HMF rehydration instead of via FAL. The
data shown in Scheme 5 also suggest that 25% reaction flux
flows through FAL, which contributes to the excess of FA that
is predicted by the microkinetic model (colored lines) in
Figure 5. Our kinetic experiments at 1408C and 0.1 molLꢀ1 HCl
revealed that after 2 h of reaction (glucose conversion ~15%)
the FA/LA ratio is close to 1 (Figure 5; solid points). This implies
that under this particular reaction condition, the reaction path
towards FAL is probably not active, which indicates that most
of the FA and LA is produced through HMF rehydration instead
of via FAL. To investigate this further, we conducted NMR spec-
troscopy with 13C-labeled glucose. The 13C NMR spectra of reac-
tion mixtures after 2 h of reaction are shown in Figure 7. As
The NMR spectroscopy and kinetic studies suggest strongly
that the path through intermediate A is less important, where-
as our MKM simulation results overestimate the contribution
from this reaction pathway. To assess our model and the im-
portance of this pathway, we have performed uncertainty anal-
ysis. If the activation free energy of the first dehydration of
protonated glucose—the rate-limiting step— is varied, the re-
action path through intermediate A can be shut down com-
pletely (Figure S8). The free energy variation range for the un-
certainty analysis was less than 5.0 kcalmolꢀ1. The uncertainty
analysis also indicates that the perturbation of the activation
energy of elementary steps that are not kinetically significant
does not result in such large variations of the reaction fluxes.
As the mean error of the B3LYP functional can be as large as
7.0 kcalmolꢀ1,[15] the mechanism proposed here can capture
the correct chemistry observed in experiments by adjusting
the reaction data of the rate-limiting step within the error of
the DFT calculations. Given the small contribution from the re-
action pathway via intermediate A, shutting down this path
completely does not significantly affect the overall re-
action activity and selectivity (Figure S9).
In addition to the reaction pathway that generates
FA with the C1 carbon atom of glucose, there is an-
other reaction channel that produces FA with a la-
1
beled C6 carbon atom. H NMR spectra of the prod-
uct mixtures after 16 h of reaction using unlabeled,
13C1-, and 13C6-labeled glucose under the same reac-
tion conditions as those used to generate the data
shown in Figure 5 are presented in Figure S14. If un-
1
labeled glucose is used, the H NMR spectrum shows
a peak at approximately d=8.05 ppm, which is char-
acteristic of FA. In the case of 13C1-labeled glucose,
two FA peaks at d=8.35 and 7.79 ppm are ascribed
Figure 7. 13C NMR spectra of the product mixtures after 2 h of reaction using 13C-labeled
glucose and the corresponding labeled products. Conditions are the same as those
stated in Figure 5.
1
to the coupling between the H atom and an adjoin-
ing 13C atom (13C NMR satellites). This result confirms
that the FA is labeled with 13C. However, the H NMR
spectrum of the reaction mixture with 13C1-labeled
a result of the low natural abundance of 13C, reactions were
performed using 13C-enriched glucose at different positions in
an effort to follow each carbon atom to the final product.
glucose revealed that there is an additional peak at d=
8.05 ppm attributed to unlabeled FA. This fraction of unlabeled
FA is ~20% of the total FA, which indicates that this pathway
is relatively minor, especially at low conversions, short times, or
low temperatures. 1H NMR spectroscopy indicates that the
main carbon atom that contributes to this additional FA comes
from the 13C6 carbon atom of glucose. Pathways generated by
the rule input network generator (RING; Figure S10B) suggest
that this path is active through a retro-aldol reaction mecha-
nism. As a retro-aldol reaction usually has a high activation
energy and is thermodynamically uphill,[7a] this alternative
pathway probably opens up at higher temperatures and
higher glucose conversion. Given the relatively minor contribu-
tion of this pathway, we have not explored it any further by
electronic structure calculations.
1
13C NMR and H NMR spectra of glucose, HMF, FA, and LA solu-
tions were also measured as standards (Figures S11 and S12).
The NMR results (complete labeling is shown in Figure 7) show
that during the dehydration reaction, the C1 atom of glucose
ends up at the carbonyl group of HMF, which implies that the
reaction pathway via intermediate B dominates over that via
intermediate A. Additionally, the characteristic resonance of FA
at d=165.5 ppm indicates that after HMF rehydration, the C1
carbon atom ends up in FA; the rest of the carbon atoms are
found in LA. To rationalize this and to understand if the path
via intermediate A is active, we extended the reaction time to
16 h (50% glucose conversion), under conditions identical to
those used to generate the data shown in Figure 5. Notably,
FA that carries the labeled C3 carbon atom of glucose was ob-
served after 16 h of reaction (Figure S13), which indicates that,
at higher conversion, the alternative pathway to FAL via inter-
mediate A opens up.
Conclusions
A DFT-based microkinetic model for a comprehensive reaction
network was developed to examine Brønsted acid-catalyzed
&
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
6
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
glucose dehydration. The DFT calculations show that, in addi-
tion to the glucose dehydration mechanism to 5-hydroxylme-
thylfurfural (HMF) hypothesized previously, direct dehydration
without HMF formation is also plausible for both glucose and
fructose in acidic solutions. The apparent activation free ener-
gies of the two dehydration pathways are consistent with ex-
perimental observations. We have shown that furfuryl alcohol
formation is kinetically accessible during the glucose dehydra-
tion and that the isomerization of glucose to fructose appears
to have a negligible contribution to the overall rate of glucose
consumption.
DFT calculations
DFT calculations were performed by using Gaussian 09.[16] Geome-
try optimizations of all the solution-phase species were performed
at the B3LYP[17]/6–31+G(d,p)[18] level of theory with a CPCM[19] treat-
ment of the water solvent using Bondi radii. In addition, two ex-
plicit water molecules were included in the reaction system in the
vicinity of the reacting centers. Stable intermediate states were
confirmed by the absence of imaginary frequencies. Transition
states were obtained by using the Berny algorithm and verified by
the presence of a single imaginary frequency. The determination of
transition state geometries to intermediate A and intermediate B
was followed by intrinsic reaction coordinate (IRC) calculations.
The microkinetic model shows that the dehydration rate de-
pends strongly on the reaction temperature and pH, in excel-
lent agreement with the experimental results. The model also
reveals that the rate-controlling step is the first dehydration of
protonated glucose. The reaction network analysis indicates
that the cyclic pathway is predominant over the acyclic one.
Further analysis of elementary reaction suggests that the reac-
tion pathway through HMF dominates at low temperatures or
low glucose conversions, whereas at high temperatures or
high glucose conversions, reaction pathways that involve
retro-aldol and secondary furfuryl alcohol chemistry may con-
tribute to the excess of formic acid observed in the experi-
ments.
To calculate the absolute free energy of a H2O molecule in water,
we need to calculate the “self-solvation free energy”, DGself*, which
is the free energy associated with moving a gas-phase molecule at
1 molLꢀ1 to the pure liquid phase. DGself* of H2O is ꢀ6.318 kcal
molꢀ1 as calculated using Equation (3):[20]
0
DGself ¼ ꢀ2:303 RT logððMliq=M Þ=ðPvapor=P0ÞÞ
ð3Þ
*
in which M0 is equal to 1 molLꢀ1, P0 is the pressure (24.47 bar) of
an ideal gas at 1m concentration and 298.15 K, and Mliq is
55.34 molLꢀ1, which is the molarity of H2O in its pure liquid form.
The absolute free energy of H2O in the liquid phase is then equal
to [Eq. (4)]:
0
*
*
*
G
ðliqÞ
¼ G þ DG0! þ DGself
ð4Þ
ðgÞ
Methodologically, we show how isotopic tracing and kinetic
experiments can be combined with elementary reaction flux
analysis and automatic mechanism generation to provide
a quantitative way to assess the mechanism and to propose
unknown pathways among numerous ones that could be stud-
ied by first principles methods. Furthermore, we illustrate how
uncertainty quantification is essential to close the gap between
experimental data and modeling. Specifically, we show that
the excess formic acid (compared to levulinic acid) seen experi-
mentally under certain conditions does not form mainly
through the furfuryl alcohol pathway but rather by a sequence
of reactions that entail a retro-aldol pathway.
0
in which G(g) denotes the gas-phase free energy. The standard
state correction (DG0!*=1.894 kcalmolꢀ1) is also included and is
caused by the change from the standard state of an ideal gas in
the gas phase (1 bar) to the experimental standard state in solu-
tion (1.0 molLꢀ1).
Microkinetic modeling
Brønsted acid-catalyzed glucose dehydration proceeds through
a complex network of elementary steps and also involves numer-
ous side reactions that lead to undesired byproducts such as
humins. The crucial ingredients of a microkinetic model are the
identification of all relevant elementary processes and the determi-
nation of the associated rate constants. The rate constant of each
elementary step was calculated using the Eyring equation [Eq. (5)]:
Experimental Section
!
Experimental methods
z
kBT
2pꢀh
ꢀDG
ðM1ꢀm
Þ
ð5Þ
k ¼
exp
kBT
Glucose 99% (Sigma–Aldrich), HCl (0.1m), and distilled water were
mixed in appropriate concentrations and used for solution prepara-
tion. Glucose labeled with 13C at different positions was purchased
from Cambridge Isotope Laboratories and used without further
treatment. Glucose dehydration kinetic experiments were per-
formed using 10-mL-thick-walled glass vials heated at 1408C in
a temperature-controlled oil bath. Vials with reaction mixture were
removed periodically and quenched in an ice bath. The reaction
mixture was filtered through a 0.2 mm filter before collecting
a sample for analysis. The liquid samples were analyzed by HPLC
(Waters e2695 with an Aminex HPX-87H column and 2414 refrac-
tive index detector).
in which kB is the Boltzmann constant, T is the temperature, hꢀ is
the Plank constant, DG° is the activation free energy, M is the mo-
larity, and m is the molecularity of the reaction. The DFT-calculated
kinetic parameters are listed in Table S1.
Kinetic parameters associated with the cyclic pathway for the con-
version of glucose to FAL were obtained from DFT calculations pre-
sented in this article. For the acyclic mechanism, which proceeds
though fructose, we employed kinetic data from earlier computa-
tional studies.[2a] The kinetic parameters for the side reactions were
estimated from experimental data and are listed in Table S2. Specif-
ically, the rates of humins formation from glucose and HMF and
the rate of HMF degradation to LA and FA are based on kinetic
data published by Weingarten et al.,[13] the rate of humins forma-
tion from fructose is modeled using kinetic data by Asghari and
Yoshida[21] and a recent microkinetic study by Nikbin et al.[2b] The
1
13C NMR and H NMR spectra (acquired with 1024 and 256 scans,
respectively) were collected by using an AV400 NMR spectrometer
equipped with a cryogenic QNP probe. Liquid samples (0.1 mL)
after reaction were mixed with D2O (1.5 mL) in 5 mm NMR
400 MHz tubes.
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
7
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
X. F. Wei, J. Phys. Chem. B 2012, 116, 10898–10904; f) Q. Xiang, Y. Y. Lee,
R. W. Torget, Appl. Biochem. Biotechnol. 2004, 115, 1127–1138; g) G.
Yang, E. A. Pidko, E. J. M. Hensen, J. Catal. 2012, 295, 122–132.
[8] a) M. J. Antal, T. Leesomboon, W. S. Mok, G. N. Richards, Carbohydr. Res.
1991, 217, 71–85; b) M. S. Feather, Tetrahedron Lett. 1970, 11, 4143–
4145; c) D. W. Harris, M. S. Feather, Carbohydr. Res. 1973, 30, 359–365;
d) D. W. Harris, M. S. Feather, J. Am. Chem. Soc. 1975, 97, 178–182;
e) D. W. Harris, M. S. Feather, J. Org. Chem. 1974, 39, 724–725; f) H.
Jadhav, C. M. Pedersen, T. Soiling, M. Bols, ChemSusChem 2011, 4,
1049–1051.
[9] M. J. Antal, W. S. L. Mok, G. N. Richards, Carbohydr. Res. 1990, 199, 91–
109.
[10] G. M. Gonzꢁlez Maldonado, R. S. Assary, J. A. Dumesic, L. A. Curtiss,
Energy Environ. Sci. 2012, 5, 8990–8997.
[11] G. M. Gonzꢁlez Maldonado, R. S. Assary, J. Dumesic, L. A. Curtiss, Energy
Environ. Sci. 2012, 5, 6981–6989.
rate of FAL rehydration to LA is taken from a kinetic study by
Hronec et al published recently.[22]
Identification of the FA formation pathways
Based on experimental evidence,[13] a minor reaction channel that
forms FA, other than through HMF hydrolysis, is observed under
certain conditions. To find possible alternative pathways to FA—in
addition to the pathway identified by DFT—we employed the
RING[23] software, and by applying a set of elementary reaction
rules, specified in the Supporting Information, we generated sever-
al reaction pathways to FA (Figure S10). These potential pathways
have been tested in juxtaposition with isotopic tracing experi-
ments.
[12] H. M. Pilath, M. R. Nimlos, A. Mittal, M. E. Himmel, D. K. Johnson, J. Agric.
Food Chem. 2010, 58, 6131–6140.
Acknowledgements
[13] R. Weingarten, J. Cho, R. Xing, W. C. Conner, G. W. Huber, ChemSusChem
2012, 5, 1280–1290.
[14] M. Salciccioli, M. Stamatakis, S. Caratzoulas, D. G. Vlachos, Chem. Eng.
Sci. 2011, 66, 4319–4355.
[15] a) S. Grimme, C. Muck-Lichtenfeld, E. U. Wurthwein, A. W. Ehlers, T. P. M.
Goumans, K. Lammertsma, J. Phys. Chem. A 2006, 110, 2583–2586;
b) E. R. Johnson, P. Mori-Sanchez, A. J. Cohen, W. T. Yang, J. Chem. Phys.
2008, 129, 204112.
[16] R. A. Gaussian 09, M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria,
M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone, B. Mennucci, G. A.
Petersson, H. Nakatsuji, M. Caricato, X. Li, H. P. Hratchian, A. F. Izmaylov,
J. Bloino, G. Zheng, J. L. Sonnenberg, M. Hada, M. Ehara, K. Toyota, R.
Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao, H.
Nakai, T. Vreven, J. A. Montgomery Jr., J. E. Peralta, F. Ogliaro, M. Bear-
park, J. J. Heyd, E. Brothers, K. N. Kudin, V. N. Staroverov, R. Kobayashi, J.
Normand, K. Raghavachari, A. Rendell, J. C. Burant, S. S. Iyengar, J.
Tomasi, M. Cossi, N. Rega, N. J. Millam, M. Klene, J. E. Knox, J. B. Cross, V.
Bakken, C. Adamo, J. Jaramillo, R. Gomperts, R. E. Stratmann, O. Yazyev,
A. J. Austin, R. Cammi, C. Pomelli, J. W. Ochterski, R. L. Martin, K. Moro-
kuma, V. G. Zakrzewski, G. A. Voth, P. Salvador, J. J. Dannenberg, S. Dap-
prich, A. D. Daniels, ꢂ. Farkas, J. B. Foresman, J. V. Ortiz, J. Cioslowski,
D. J. Fox, Gaussian, Inc., Wallingford CT, 2009.
The theoretical work was conducted by L.Y. and was supported
by the US Department of Energy, Energy Efficiency and Renewa-
ble Energy’s Bioenergy Technologies Office with Pacific Northwest
National Laboratory (PNNL). PPNL is operated by Battelle for the
US Department of Energy. The experimental work was conducted
by G.T. and was supported by the Catalysis Center for Energy In-
novation, an Energy Frontier Research Centre funded by the U. S.
Department of Energy, Office of Science and Office of Basic
Energy Sciences under award number DE-SC0001004.
Keywords: ab initio calculations · kinetics · isotopic labeling ·
NMR spectroscopy · reaction mechanisms
[1] a) D. M. Alonso, J. Q. Bond, J. A. Dumesic, Green Chem. 2010, 12, 1493–
1513; b) M. Besson, P. Gallezot, C. Pinel, Chem. Rev. 2014, 114, 1827–
1870; c) P. Gallezot, Chem. Soc. Rev. 2012, 41, 1538–1558.
[2] a) S. Caratzoulas, D. G. Vlachos, Carbohydr. Res. 2011, 346, 664–672;
b) N. Nikbin, S. Caratzoulas, D. G. Vlachos, ChemCatChem 2012, 4, 504–
511.
[3] a) J. S. Kruger, V. Nikolakis, D. G. Vlachos, Curr. Opin. Chem. Eng. 2012, 1,
312–320; b) J. S. Kruger, V. Nikolakis, D. G. Vlachos, Appl. Catal. A 2014,
469, 116–123; c) J. C. Serrano-Ruiz, R. M. West, J. A. Dumesic, Annu. Rev.
Chem. Biomol. 2010, 1, 79–100; d) D. A. Simonetti, J. A. Dumesic, Catal.
Rev. 2009, 51, 441–484.
[4] B. Girisuta, L. P. B. M. Janssen, H. J. Heeres, Chem. Eng. Res. Des. 2006,
84, 339–349.
[5] a) E. Nikolla, Y. Roman-Leshkov, M. Moliner, M. E. Davis, ACS Catal. 2011,
1, 408–410; b) V. Choudhary, S. H. Mushrif, C. Ho, A. Anderko, V. Nikola-
kis, N. S. Marinkovic, A. I. Frenkel, S. I. Sandler, D. G. Vlachos, J. Am.
Chem. Soc. 2013, 135, 3997–4006.
[17] A. D. Becke, J. Chem. Phys. 1993, 98, 5648–5652.
[18] a) R. Ditchfield, W. J. Hehre, J. A. Pople, J. Chem. Phys. 1971, 54, 724–
728; b) W. J. Hehre, R. Ditchfield, J. A. Pople, J. Chem. Phys. 1972, 56,
2257–2261; c) M. M. Francl, W. J. Pietro, W. J. Hehre, J. S. Binkley, M. S.
Gordon, D. J. Defrees, J. A. Pople, J. Chem. Phys. 1982, 77, 3654–3665.
[19] a) V. Barone, M. Cossi, J. Phys. Chem. A 1998, 102, 1995–2001; b) M.
Cossi, N. Rega, G. Scalmani, V. Barone, J. Comput. Chem. 2003, 24, 669–
681; c) A. Klamt, G. Schuurmann, J. Chem. Soc. Perkin Trans. 1993, 2,
799–805.
[20] C. P. Kelly, C. J. Cramer, D. G. Truhlar, J. Phys. Chem. B 2007, 111, 408–
422.
[21] F. S. Asghari, H. Yoshida, Ind. Eng. Chem. Res. 2007, 46, 7703–7710.
[22] M. Hronec, K. Fulajtarova, T. Sotak, J. Ind. Eng. Chem. 2014, 20, 650–655.
[23] a) S. Rangarajan, A. Bhan, P. Daoutidis, Comput. Chem. Eng. 2012, 46,
141–152; b) S. Rangarajan, A. Bhan, P. Daoutidis, Comput. Chem. Eng.
2012, 45, 114–123.
[6] R. S. Assary, P. C. Redfern, J. Greeley, L. A. Curtiss, J. Phys. Chem. B 2011,
115, 4341–4349.
[7] a) R. S. Assary, L. A. Curtiss, Energ. Fuel 2012, 26, 1344–1352; b) R. S.
Assary, T. Kim, J. J. Low, J. Greeley, L. A. Curtiss, Phys. Chem. Chem. Phys.
2012, 14, 16603–16611; c) X. H. Qian, J. Phys. Chem. A 2011, 115,
11740–11748; d) X. H. Qian, Top. Catal. 2012, 55, 218–226; e) X. H. Qian,
Received: November 13, 2014
Published online on && &&, 0000
&
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
8
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

FULL PAPERS
Glucose dehydration: The DFT-based
microkinetic modeling of the Brønsted
acid-catalyzed dehydration of glucose is
in excellent agreement with experimen-
tal kinetic data. The rate-limiting step is
the first dehydration of protonated glu-
cose. Isotopic tracing NMR spectroscopy
reveals that glucose dehydrates through
a cyclic path and the majority of glu-
cose is consumed by the 5-hydroxyme-
thylfurfural intermediate at low conver-
sions.
L. Yang, G. Tsilomelekis, S. Caratzoulas,
D. G. Vlachos*
&& – &&
Mechanism of Brønsted Acid-
Catalyzed Glucose Dehydration
ChemSusChem 0000, 00, 0 – 0
www.chemsuschem.org
9
ꢀ 0000 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ