Recent Advances in Photocatalytic Transformation of Carbohydrates Into Valuable Platform Chemicals

In response to the less accessible fossil resources and deteriorating environmental problems, catalytic conversion of the abundant and renewable lignocellulosic biomass to replace fossil resources for the production of value-added chemicals and fuels is of great importance. Depolymerization of carbohydrate and its derivatives can obtain a series of C5-C6 monosaccharides (e.g., glucose and xylose) and their derived platform compounds (e.g., HMF and furfural). Selective transformation of lignocellulose using sustainable solar energy via photocatalysis has attract broad interest from a growing scientific community. The unique photogenerated reactive species (e.g., h+, e−, •OH, •O2 −, and 1O2), novel reaction pathways as well as the mild reaction conditions make photocatalysis a “dream reaction.” This review is aimed to provide an overview of the up-to-date contributions achieved in the selective photocatalytic transformation of carbohydrate and its derivatives. Photocatalytic methods, properties and merits of different catalytic systems are well summarized. We then put forward future perspective and challenges in this field.


INTRODUCTION
As a result of the increase in world population and rapid industrial development over the past century, the world has become increasingly dependent on nonrenewable fossil resources such as petroleum, coal to meet the growing demand and the increasingly challenging industrial and transportation technology requirements (Puga, 2016). Nowadays, about 85% of all fossil feedstocks consumed is used for the production of transportation fuels and only 10% for the production of chemicals (Climent et al., 2011). Against the backdrop of predicted depletion of worldwide mineral resources and growing concerns about political and environmental concerns such as global warming and deteriorating air quality, the focus of research is shifting to alternative and sustainable energy resources to produce inexpensive and transportable fuels that can be directly employed in current power generation systems (De et al., 2015;Hu et al., 2017). The production of carbon-neutral, lowemission biofuels and green chemicals from renewable resources, such as biomass, is particularly compelling in the progressive replacement of traditional fossils.
Biomass is the world's fourth largest energy source after oil, coal and natural gas (Li et al., 2015). Photosynthesis in nature produces 170 billion metric tons of biomass per year by fixing carbon from atmospheric CO 2 . . Plant material consisting mainly of carbohydrate polymers (cellulose and hemicellulose) and an aromatic polymer (lignin) is considered to be lignocellulose, which represents more than 90% of all plant biomass. The represented structure of lignocellulose is illustrated in Figure 1. Cellulose and hemicellulose are encapsulated in a lignin shelter. Cellulose fibers bind to hemicellulose and lignin by hydrogen bonding, while hemicellulose is covalently bound to lignin. The percentage of each component varies depending on the plant species, but generally, typical lignocellulose is composed of 35-50% cellulose, 25-35% hemicellulose, and 10-25% lignin . Carbohydrates, in particular, are promising as they are the largest source of natural renewable carbon by far (van Putten et al., 2013), and harbor great potential to produce a variety of value-added commodity chemicals (e.g., gluconic acid, 5hydroxymethyl furfural (HMF), 2,5-furandicarboxylic acid (FDCA), levulinic acid (LA)) (Sheldon, 2014). Surprisingly, less than 5% of these biomass carbohydrates have so far been used by humans for a variety of purposes (Zhang et al., 2017b). A large portion of them is considered to be waste streams and remain underutilized at present. Their fate is burned to provide heat and energy for biorefinery and paper/pulp industries. It is worth noting that the use of vast amount of edible biomass as long-term feedstocks for the production of first-generation biofuels (biodiesel and bio-ethanol) require significant land input, thus posing a direct negative impact on global food supply. So recent researches have focused on the valorization of non-edible biomass or waste biomass generated during the production of edible crops (Sheldon, 2014). According to the International Energy Agency (IEA), biofuels could provide 10% of the world's primary energy supply by 2035, and replace 27% of the world's transportation fuel by 2050 . Thus, holistic utilization of the world's most abundant bio-resources could greatly alleviate energy crisis.
In traditional lignocellulosic biorefineries, thermochemical methods (i.e., pyrolysis, gasification and liquefaction) have been developed to transform carbohydrates into bio-oil, syngas or liquid fuels at high temperature and/or pressure, resulting in intensive thermal energy input. Biochemical treatment, including digestion and fermentation, is another useful method to convert biomass into biogas using costly peculiar enzymes, bacterial and microorganisms under anaerobic conditions. In contrast, Sun light as an inexhaustible energy source can be used to replace thermal energy. Photocatalysis, in which photons are used to drive redox reactions, represents a promising strategy to convert solar energy and renewable lignocellulose into valuable fuels and chemicals. The active photogenerated holes and electrons, as well as unique photo-induced reactive oxygen species (ROS) enable the precise cleavage of targeted chemical bonds. In addition, various functional groups can be retained under mild reaction conditions (Zhang, 2018).
A growing number of research articles dealing with photocatalytic conversion of carbohydrates (i.e., cellulose, glucose, HMF) into fuels and chemicals have been published. However, these contributions have not been comprehensively reviewed previously (Liu et al., 2019;Wu et al., 2020). In order to promote further development in such a promising area, there is a need for a timely and systematic review of up-to-date advances in this area. Therefore, this review is aimed to provide an overview of the state-of-the-art contributions achieved along the lines of selective photocatalytic transformation of carbohydrate and its derivatives. Future perspective and challenges in this field will also be emphasized. The authors hope that this review can provide guidance and inspiration to develop efficient photocatalyst for the valorization of carbohydrates.
fiber. Cellulose, with a general formula of (C 6 H 12 O 5 ) n , possesses the simplest structure among other polysaccharides, as it is a homopolymer composed of glucose in the form of β-Danhydroglucopyranose units (AGUs) that covalently linked to each other by β-1,4-glycosidic bonds between the equatorial OH groups of C 4 and the C 1 carbon atoms (Credou and Berthelot, 2014). The number of repeating AGUs constituting the polymer chain is defined as the degree of polymerization (DP) of cellulose. For instance, the common DP value of cellulose derived from wood pulp is around 300-1700, whereas the value is about 800-10,000 in cotton or other plant fibers (Klemm et al., 2005). Large number of hydroxyl groups along the skeleton form intra-and inter-molecular hydrogen bonds with oxygen atoms on the same or another chain, holding the chains firmly to form microfibrils with high tensile strength. Thus, dissolution of cellulose without chemical modification in common solvents is rather difficult to achieve (Hu et al., 2015;Shaghaleh et al., 2018). It was reported that this hydrogen-bonding network could be broken with concentrated ZnCl 2 owing to the interaction between ionic species and hydroxyl groups (Fan et al., 2011).
The selective transformation of cellulose into various highvalue bio-based chemicals and high-quality fuels has drawn intensive attention in the field of biorefinery. It should be pointed out that hydrolytic cleavage of the β-1,4-glycosidic bonds between two anhydroglucose units to produce glucose is the starting point for other catalytic transformations (Huang and Fu, 2013). Owing to its crystalline structure and the robust intra-and intermolecular hydrogen bonding interaction, hydrolysis of cellulose is significantly challenging. More recently, it was found that localized surface plasmon resonance (LSPR) effect has a potential in facilitate biomass conversion. Compared with traditional heating method, LSPRinduced photothermal heating is more effective and straightforward toward intended reaction sites. It was first reported that TiO 2 nanofibres supported H-form Y-zeolites (HY) decorated with Au nanoparticles (denoted as Au-HYT) showed an enhanced ability in cellulose hydrolysis under visible light irradiation . Pores on nanozeolites were elaborately controlled to facilitate mass transfer of dissolved cellulose into the porous channel to interact with the active site. The decorated plasmonic Au NPs absorb visible light and enhance the polarized electric field of zeolites, which also contributes to the increase in acid strength. Both glucose and HMF were detected as main hydrolysis products, with a total yield of more than 60% at 130°C for 24 h. The yield of glucose increased with an increase in light intensity, indicating that the conversion of cellulose is truly enhanced by LSPR effect.
Also based on the LSPR effect, Ir/HY zeolite catalyst exhibited a different mechanism for the hydrolysis of cellobiose, a dimer of glucose with β-1,4-glycosidic bonds (Zhang et al., 2018a). The role of Ir NPs and HY zeolite were carefully clarified. Under visible light irradiation, the highest temperature of Ir/HY reached 106°C, while the temperature of HY zeolite without Ir NPs was only about 40°C. This result demonstrated that the plasmonic photothermal effect of Ir NPs effectively converts light into thermal energy, increasing the local temperature of the surface of Ir/HY catalysts, thus accelerating the catalytic hydrolysis of cellobiose. Meanwhile, the acidic sites of HY zeolite as active centers are responsible for the hydrolysis of cellobiose. The role of the acid strength of the solid acid catalysts in activating crystalline cellulose toward hydrolysis was further explored. Generally, HY zeolites with higher ratios of Si/Al have stronger acidity. The observed activity is in the order of Ir/HY 3 (Si/Al 11)> Ir/HY (Si/ Al 7)> Ir/HY 2 (Si/Al 5.2), suggesting that the acid density of the solid acid catalyst is directly correlated with catalytic activity. Only glucose and HMF were detected, which were derived from the breakage of β-1,4-glycosidic bonds, not from the C−C bonds cleavage, indicating very high selectivity (>99%). This is in accordance with the results reported by Wang . Ir/HY 3 and Ir/Amberlyst-15 resin (a strong commercial protonic solid acid) catalysts were further applied to treat nonpretreated crystalline cellulose. Under visible light irradiation, the yield of atotal products was up to 75.3% at 90°C and 72.6% at 70°C, respectively.
Although great progresses have been made in photoenhanced solid acid-catalyzed hydrolysis of cellulose, mass diffusion between crystalline cellulose and solid acid catalyst still impedes the efficient cellulose transformation. Concurrent modification of TiO 2 photocatalyst with chemisorbed sulfate (SO 4 2−-) and nickel sulfide (Ni x S y ) endows TiO 2 with another identity as a solid acid catalyst (Hao et al., 2018). The SO 4 2species bound to the TiO 2 surface by coordination facilitate cellulose hydrolysis into glucose which is more attainable to catalyst ( Figure 2). Photogenerated holes oxidize the water to generate • OH radicals, which attack glucose to produce protons and formic intermediate. Meanwhile, Ni x S y cocatalyst favors charge separation and act as hydrogen evolution sites for H 2 generation. However, the accumulated formate may adsorb on the catalyst surface and occupy the corresponding active sites to decrease the H 2 evolution rates. A significant elevation in H 2 yield was achieved using TiO 2 -SO 4 2− -Ni x S y , reaching 3.02 mmol g −1 h −1 on average during the first 3 h, which was approximately 76 times higher than that of pristine P25.
Zhang et al. developed a photocatalytic system for the photo-reforming cellulose into H 2 in neutral aqueous solution by loading a graphitic carbon layer on TiO 2 /NiO x nanoparticles (denoted as TiO 2 /NiO x @C g ) (Zhang et al., 2018b). Enhanced H 2 production yields of ∼270 μmol h −1 g −1 and ∼4,000 μmol h −1 g −1 were obtained at room temperature and 80°C, respectively. The significant enhancement of the hydrogen production rate may be because the higher temperature promotes the dissolution of a portion of suspended carbohydrate fragments, which provides solvated cellulose as substrate for photocatalysis. The mechanism of the process has been proposed in Figure 3. Upon light irradiation, the deposited NiO x is reduced to metallic Ni by the photogenerated electrons from TiO 2 . A proton is withdrawn from the −OH group of the alcohol adsorbed on the Ni surface and subsequently reduced by the electron to obtain a Ni-H hydride. The presence of the carbon layer weakens the interaction of H with the Ni metal surface, generating a molecule of H 2 . Meanwhile, alkoxide anion left behind is oxidized to CO 2 , CO, CH 4 by holes.

PHOTOCATALYTIC OXIDATION OF GLUCOSE
Glucose is the most abundant and cheapest monosaccharide in nature, which can be easily obtained by hydrolysis of polysaccharides such as cellulose and starch (Deng et al., 2014;Chatterjee et al., 2015). With six carbon atoms, glucose is classified as hexose and it can exist in an open-chain form as well as a cyclic form. In its open-chain form, the glucose molecule has an open and unbranched six-carbon backbone, where C 1 is part of an aldehyde group and each of the other five carbons bears a hydroxyl group. Therefore, glucose is also known as aldose or aldohexose. The aldehyde group makes glucose a reducing sugar giving a positive result with the Fehling solution. The cyclic form is the result of an intramolecular reaction between the aldehyde C 1 atom and the C 5 hydroxyl group to form an intramolecular hemiacetal.
Selective oxidation of glucose can produce a wide range of value-added chemicals, such as gluconic acid, glucaric acid, formic acid and so on ( Figure 4). These products as valuable platform chemicals have been widely used in many fields. For example, gluconic acid is widely used as food additive, concrete ingredient medical intermediate and precursor for biodegradable polymers (Climent et al., 2011). Glucaric acid, regarded as one of the top value-added chemicals from biomass, is a key building block for producing various functional polymers including new nylons and hyperbranched polyesters (Kiely et al., 1994). Formic acid as an important intermediate is widely used in chemical synthesis and also has the potential to serve as an excellent hydrogen carrier because the dehydrogenation of formic acid can easily proceed under mild conditions using catalysts (Guerriero et al., 2014).
Traditional oxidation of glucose has been extensively studied through fermentation and erobic catalytic oxidation (Benkó et al., 2014;Haynes et al., 2017;Liu et al., 2017). However, the fermentation process proceeds relatively slow, which is greatly depend on the enzyme activity. Aerobic oxidation using heterogeneous catalysts based on precious metal nanoparticles (i.e., Au, Ag, Pd, and Pt) was usually carried out at high temperature and pressure, resulting in a low performanceprice ratio and security risks. In contrast to these energy intensive processes, recent application of heterogeneous photocatalysts in the transformation of glucose with oxygen as the terminal oxidant has received increasing attention because photocatalytic transformation is driven by solar energy and can be conducted at room temperature and atmospheric pressure. TiO 2 -based photocatalysts are the most widely used owing to their non-toxicity, low cost and excellent stability. A powdered TiO 2 photocatalyst was synthesized by an ultrasound-promoted sol-gel method for the selective photo-oxidation of glucose to glucaric acid, gluconic acid and arabitol. Experiment results indicated that a mixture of 50:50 H 2 O: ACN (v/v) was the best solvent composition. After 10 min of irradiation, glucose conversion reached 11% and total organic selectivity was up to 71.3%. The relatively high carboxylic acids selectivity could be attributed to lower affinity of these acids on TiO 2 (US) surface in the presence of acetonitrile as "stabilizing agent". Meanwhile, lower amount of water might give rise to a lower concentration of nonselective • OH radicals (Colmenares et al., 2011).
A more selective zeolite Y-supported TiO 2 photocatalyst was fabricated, offering total selectivity of gluconic and glucaric acid FIGURE 2 | Proposed mechanism of photo-reformation of cellulose into H 2 using TiO 2 -SO 4 2--Ni x S y .
FIGURE 3 | Proposed potential mechanism of the photoreforming of cellulose by TiO 2 /NiO x @C g .
Frontiers in Chemical Engineering | www.frontiersin.org March 2021 | Volume 3 | Article 615309 up to 68%. The electrostatic repulsion between negatively charged zeolite framework and carboxylic acids facilitated the desorption of organic acids, preventing subsequent mineralization (Colmenares and Magdziarz, 2013). This kind of electrostatic repulsion also exists in the alkaline suspension of nano-TiO 2 . A formate yield of 35% and high conversion of glucose up to 100% at ambient condition with 0.03 M NaOH was achieved. The precise control of the amount NaOH is fundamental in the conversion of glucose because appropriate amount of hydroxyl ions can not only promote the formation of active oxidative radicals ( • O 2 − , • OH), but also make the surface of TiO 2 negatively charged. The repulsive forces between formate ions (HCO 2 − ) and TiO 2 surface would facilitate the desorption of formic acid and then a high formate selectivity can be obtained .
The incorporation of Cr 3+ or Fe 3+ ions into TiO 2 lattice would slightly reduce the bandgap to the value of 3.04 or 2.3 eV, compared with generally accepted 3.2 eV for anatase TiO 2 . A total organic acid selectivity of 87% at a 7% glucose conversion was obtained for Cr-doped TiO 2 /zeolite, and the Fe-modified TiO 2 /zeolite offered a 94.3% selectivity to glucaric and gluconic acid (Colmenares et al., 2013a;Colmenares et al., 2013b). The effect of co-doping on the physicochemical properties of TiO 2 for the conversion of glucose into value-added chemicals were rarely reported. When TiO 2 is co-doped by boron (B) and nitrogen (N), the introduction of foreign B and N element results in the formation of localized states near the CB and VB in the bandgap of TiO 2 , respectively. Excitations from these localized states to the CB of TiO 2 may be responsible for the red shift of the absorption edge toward the visible region. For Ag/N-doped TiO 2 catalyst, Ag acts as electron traps to facilitate charge separation. The improved charge separation and increased active surface area lead to enhanced generation of hydroxyl radicals, which further oxidize glucose to various high-value chemicals (Suriyachai et al., 2020).
In evaluating the performance of a photocatalyst, we typically focus on the ease of electron-hole pair separation, charge mobility and the rate at which the active species is produced. However, the interaction between substrate and photocatalyst surface is sometimes overlooked, but this is exactly what is crucial and may provide new understanding of the reaction pattern. Visiblelight selective photocatalytic conversion of glucose using unmodified TiO 2 is achieved through the formation of a glucose-TiO 2 charge transfer complex (Da Vià et al., 2017). The formation of this kind metal-organic complex responds for the visible-light absorption ability of the TiO 2 and allows the direct transfer of photo-excited electrons from the highest occupied molecular orbital of glucose to the conduction band of the TiO 2 (Kim et al., 2015). Optimization of the reaction conditions resulted in 42% glucose conversion under visible light with 7% selectivity to gluconic acid and 93% selectivity to other partial oxidation products.
Han et al. reported that TiO 2 supported Au nanoparticles was very efficient and selective in glucose oxidation under both UV and visible light in Na 2 CO 3 aqueous solution .  Both glucose conversion and gluconic acid yield were up to 99% under visible light irradiation for 4 h. It is interesting that the electron generation and transfer in Au NPs/TiO 2 is completely different under visible and UV light irradiation ( Figure 5). Under visible light, electrons induced by LSPR effect were released from Au NPs and injected into the conduction band of TiO 2 . Then, these photoactivated electrons activated O 2 to form active superoxide radical ( • O 2 − ), which further participated in the oxidation of glucose. When Au NPs/TiO 2 was irradiated by UV light, photogenerated holes were excited to the conduction band of TiO 2 from the valence band. Au NPs trapped the electrons and served as reaction sites for the generation of • O 2 − , which oxidized the glucose into gluconic acid. Meanwhile, Reactive oxygen species (e.g., • OH and 1 O 2 ) with strong oxidization power were inhibited by Na 2 CO 3 , leading to a high selectivity of desired product.
Inspired by natural enzymes that exhibit both high activity and selectivity in bioreactions, a strategy of mimicking enzyme center to prepare novel biomimetic catalyst for selective oxidation reactions has been developed. Metallothioporphyrazines (MPz), a sulfur-containing macrocyclic compound with extensive conjugated system of delocalized π-electrons and strong visible light absorption ability, has been considered as one of the most promising biomimetic photocatalysts (Zhou et al., 2016;. Under visible light irradiation, MPzs can activate hydrogen peroxide or oxygen for the selective organic transformation and pollutant degradation. Several kinds of metallothioporphyrazines with different supports were prepared and further applied to the photocatalytic oxidation of glucose, such as H-ZSM-5supported FePz (SBu) 8 (Chen et al., 2019), SnO 2 -supported FePz (SBu) 8 , TiO 2 /HPW/CoPz (Yin et al., 2020), and ZnO-supported CoPzS 8 (Cheng et al., 2019). For example, a TiO 2 /HPW/CoPz biomimetic photocatalyst was synthesized by modifying TiO 2 with HPW and CoPz (Yin et al., 2020). Glucose was oxidized into gluconic acid and glucaric acid with molecular oxygen in only water without the addition of H 2 O 2 . Under optimized conditions, total selectivity for both acids reached up to 80.4% at a 22.2% glucose conversion. It is interesting that HPW and CoPz had synergistic effect on glucose oxidation. On one hand, the presence of CoPz not only improved the separation of photogenerated charges and accelerated the formation of active species but also increased the adsorption capacity of glucose on catalyst surface, resulting in a higher glucose conversion. On the other hand, the introduction of HPW increased the surface acidity of the catalyst. The repulsive force facilitates the desorption of carboxylic acids (glucaric and gluconic acids) from the catalyst surface. Therefore, further oxidation was inhibited and a higher selectivity to both acids was obtained.

PHOTOCATALYTIC TRANSFORMATION OF HMF AND FURFURAL
5-hydroxymethylfurfural (HMF) and furfural, the dehydration product of C 5 and C 6 carbohydrates, are deemed as versatile intermediates in biomass conversion and key platform compounds for the production of value-added industrial and pharmaceutical chemicals. (Zakrzewska et al., 2011;Hu et al., 2017). The marvelous and reactive structure containing an furan ring and an aldehyde group allow HMF to undergo a variety of reactions, such as oxidation, hydrogenation, etherification et al., to produce high-quality biofuels such as 2,5-dimethylfuran (DMF), 2,5-dimethyltetrahydrofuran (DMTHF), 5ethoxymethylfurfural (EMF) and new-fashioned high-value chemicals such as levulinic acid (LA), 2,5-diformylfuran (DFF), 2,5-dihydroxymethylfuran (DHMF) and 2,5furandicarboxylic acid (FDCA) (Figure 6) (Hu et al., 2018). Among them, much attention has been paid to DFF and FDCA. DFF can be widely used in the production of furanbased polymers, organic conductors, and intermediates of pharmaceuticals and antifungal agents. FDCA was regarded as one of the top 12 value-added chemicals from biomass by the United States Department of Energy in 2004, and has been deemed as an important building block for the production of biochemicals and has the potential to be a renewable alternative to terephthalic acid to produce polyethylene terephthalate (PET) plastics (Zhang and Deng, 2015).
Given that much more interest and attention have been paid to the furan-based derivatives, the state-of-the-art advances on the photocatalytic transformation of furan-based derivatives via oxidative and reductive approaches are systematically summarized and discussed ( Table 3).

Photo-Oxidation Reactions
Since the pentatomic furan ring is not as stable as that of the hexatomic aromatic structures, it is not surprising that overoxidation and/or mineralization often occurs in HMF transformation when strong oxidizing species are generated in the photocatalytic system, resulting in poor product selectivity. In 2013, HMF was reported to be photocatalytic oxidized to DFF with home-prepared TiO 2 for the first time (Yurdakal et al., 2013). TiO 2 nanoparticles with three different crystalline phases (anatase, rutile, and brookite) were prepared via a sol-gel method, but their crystallinity was far less than commercial TiO 2 catalyst. A DFF selectivity of 22% was obtained under ultraviolet (UV) irradiation. The low selectivity might be ascribed to the generation of highly oxidized • OH radicals, which unselectively attack the furan ring of HMF and mineralize them into CO 2 and H 2 O. Nitrogen doping can boost DFF selectivity to 30-40% (Krivtsov et al., 2017b). It was proposed that the N-species on the TiO 2 modified its surface chemistry by reducing the hydroxylation degree. Thus, the transfer of unselective • OH radicals from TiO 2 surface to products can be suppressed.
In comparison, g-C 3 N 4 possesses a moderate band gap of around 2.7 eV. The less positive VB edge make g-C 3 N 4 hard to oxidize water or OH − to produce unselective • OH radical by photogenerated holes. However, the more negative CB potential is in favor of activating O 2 to generate superoxide radical ion ( • O 2 − ), which is a key oxidative species for HMF oxidation. Metal-free g-C 3 N 4 was synthesized by a simple calcination of melamine. The addition of water favored the formation of pore Frontiers in Chemical Engineering | www.frontiersin.org March 2021 | Volume 3 | Article 615309 6 structure and enhanced specific surface area and pore volume. It was found that O 2 is indispensable in HMF oxidation, since it is the source of • O 2 − radical, which is determined as the dominant active species. The addition of polar solvents decreased the HMF conversion, probably because they prone to compete with the HMF for the active sites. Benzotrifluoride (PhCF 3 ) was found to be the best solvent due to its less polarity and superior ability for O 2 dissolvent .
A bulk g-C 3 N 4 catalyst was prepared via the thermal condensation method from different precursors (melamine, urea and thiourea). Only a 28% selectivity of DFF was obtained under solar light irradiation for 4 h. The thermal exfoliation procedure applied to bulk g-C 3 N 4 increased the selectivity to 42%, which can be attributed to the increased specific surface area of thermally exfoliated g-C 3 N 4 nanosheets (Krivtsov et al., 2017a). The addition of p-benzoquinone significantly inhibited the reaction, indicating that • O 2 − radicals were mainly responsible for the HMF oxidation. Changing the reaction atmosphere to nitrogen did not completely terminate the reaction, but almost no DFF was detected (Krivtsov et al., 2017a), probably due to the interaction of HMF with the surface functional groups (e.g., amino groups) of g-C 3 N 4 (Ilkaeva et al., 2018). The formation of hydrogen bond between −NH 2 groups and water allows it been attacked by the photogenerated holes to form localized • OH radicals ( Figure 7A). The reaction of HMF with • OH radicals leads to the ring opening and the formation of aliphatic intermediates, thus reducing the selectivity of DFF ( Figure 7B). However, after the reaction of g-C 3 N 4 with H 2 O 2 to give g-C 3 N 4 -H 2 O 2 adduct, H 2 O 2 blocks the amino-groups of the g-C 3 N 4 , thus creating a steric hindrance for the HMF molecule interaction with the g-C 3 N 4 surface sites ( Figure 7C). The conversion of HMF may decreased slightly due to the absence of • OH radicals, but the selectivity of DFF increased to 88% in return (Ilkaeva et al., 2018).
Frontiers in Chemical Engineering | www.frontiersin.org March 2021 | Volume 3 | Article 615309 light excitation of pyrrole rings in CoPz molecule results in excited CoPz* with higher electronic cloud density. CoPz* exhibits much higher photocatalytic activity and activates O 2 to 1 O 2 species, which possess suitable oxidation power to selectively oxidize HMF to the desired product FDCA. Then the excited CoPz* come back to its ground state ( Figure 8D). In addition, pH value of the reaction system was found to be a crucial factor in product selectivity. At pH 6.86 or 9.18 HMF was selectively oxidized to FDCA with a high selectivity above 97%. While at low pH (4.01), the selectivity of FDCA was less than 10% ( Figures 8A-C).
Analogous control of product selectivity by adjusting pH was also reported by Sun et al. . The authors found that nearly complete photocatalytic oxidation of HMF to DFF was realized by ultrathin CdS nanosheets loaded with cocatalyst Ni (Ni/CdS) under strong alkaline conditions ( Figure 8E). This can be attributed to the Cannizzaro mechanism that aldehyde groups are not stable at high pH and will disproportionate to form alcohols and acids. An interesting phenomenon was also observed that almost 100% of furfural alcohol was oxidized to furfural after 22 h photocatalysis, while only 20% HMF conversion was achieved under the same condition ( Figure 8F). DFT calculations and control experiments with DHMF and HMFCA as substrates indicated that HMF prefers to adsorb at water/NiO (001) interface via the −CHO group, impeding the oxidation of −OH group attached to the furan ring ( Figure 8G). However, the presence of ancillary −OH and −COOH group did not present any negative impact. Photoluminescence and 1 H NMR results demonstrated that nearly no • OH radicals were generated during the photocatalysis process, which is consistent with Xu's results (Xu et al., 2017). Moreover, the photocatalytic activity could be enhanced by introducing nitrate salts (e.g., lithium, magnesium, calcium, and manganese) as redox mediator (DiMeglio et al., 2019).
In recent studies, HMF was often used as a sacrificial agent in the photocatalytic H 2 evolution by water splitting. Replacing water with D 2 O, the detected D 2 indicated that the majority of H 2 is generated by the reduction water. The addition of cocatalyst Pt on the porous carbon nitride has no significant effect in the oxidation of HMF to DFF, but only contributes to the charge separation and form Pt-H bonds to reduce protons to H 2 (Battula et al., 2019). However, with NiS as cocatalyst, a Schottky barrier is formed at the interface between NiS and Zn 3 In 2 S 6 , leading to the spatial separation of photoexcited charges (Meng et al., 2020). The H 2 evolution rate and DFF production in HMF solution reached up to 120 and 129 μmol h −1 g −1 , which are 41.4 and 35.8 times higher than that of pure Zn 3 In 2 S 6 . In this work, the author confirmed the thermodynamics feasibility of the photocatalytic HMF-to-DFF oxidation and H 2 generation process. The VB potential of Zn 3 In 2 S 6 is 1.83 eV, which stands between the E HMF/DFF (1.61 eV) and E DFF/oxidized DFF (2.03 eV). This means the photogenerated holes generated on the valence band of Zn 3 In 2 S 6 can directly oxidize HMF to produce DFF, but cannot further oxidize DFF. On the other hand, the CB potential of Zn 3 In 2 S 6 (−0.97 eV) is more negative than the reductive potential of H 2 evolution (0 V). The HMF molecular is firstly deprotonated to form an alkoxide, which further react with a photoexcited hole to generate a carbon radical. The formed carbon radical is oxidized by a hole to produce a DFF molecular. Simultaneously, two protons, abstracted from HMF, are reduced by two electrons o produce a H 2 molecular.
The photocatalytic oxidation of HMF to DFF with Nb 2 O 5 under visible light was reported by Wu's group (Zhang et al., 2017a). A high DFF selectivity (90.6%) from HMF was observed using Nb 2 O 5 in benzotrifluoride in the presence of O 2 under visible light irradiation, although the bandgap energy of Nb 2 O 5 is >3.2 eV. They determined that the alcoholic hydroxyl group of HMF can be adsorbed on Nb 2 O 5 to form alcoholate species, decreasing the bandgap energy of Nb 2 O 5 to the visible range. However, the DFF yield was not very satisfactory (<20%). Compared with Nb 2 O 5 -300 and Nb 2 O 5 -500, Nb 2 O 5 treated at 800°C exhibits better photocatalytic activity, which is attributed to its high crystallinity, delaying the recombination of photogenerated charge carriers. MnO 2 nanorods could also catalyze a nearly complete conversion HMF to DFF (99% conversion, ∼100% selectivity) without any additives (bases or oxidants). The oxidation was realized through a redox cycle between Mn 4+ and Mn 3+ (Giannakoudakis et al., 2019). The detection of trace amount of Mn 2 O 3 phase on the spent catalyst  (Figure 9). However, the generated water molecules may form hydrogen bonds with the hydroxyl groups of HMF to inhibit its direct interaction with catalytic surface centers, resulting in a decreased conversion as reaction proceeded. The use of aprotic and less polar solvent will have a positive impact.
Most previous studies of HMF oxidation were carried out using noble metals in alkaline aqueous solutions at high O 2 pressure and elevated temperature. Another innovative approach to achieve HMF oxidation is electrochemical oxidation, where the oxidation capacity is driven by an applied electrochemical potential rather than by O 2 or other chemical oxidants. Furthermore, it is noteworthy if the power to supply electrochemical potential comes directly from renewable energy such as solar light, the electrocatalytic oxidation of HMF would be a more promising approach. Fortunately, this idea was realized in 2005 by constructing a photoelectrochemical cell (PEC) (Cha and Choi, 2015). A nearly 100% FDCA yield and more than 93% faradaic efficiency were obtained. In this PEC system, an n-type nanoporous BiVO 4 electrode was used as a photoanode to generate photoexcited holes for the oxidation of 2,2,6,6tetramethylpiperidine-1-oxyl (TEMPO) to TEMPO + , which was further responsible for the oxidization of HMF to FDCA in 0.5 M borate buffer solution (pH 9.2). Meanwhile, the separated electrons were subsequently transferred to Pt cathode to reduce water to H 2 . The high faradaic efficiency (FE) achieved for the FDCA formation indicated that the TEMPO oxidation is kinetically much more favorable than the water oxidation. Compared with electrochemical TEMPOmediated HMF oxidation, the onset potential necessary to initiate photooxidation of TEMPO decreased considerably. This is because the photogenerated holes in BiVO 4 already have sufficient overpotential for TEMPO oxidation before an external bias is applied, significantly decreasing the external energy input for the PEC operation ( Figure 10). The role of the applied bias is just to enhance the electron-hole pair separation to ensure that more photogenerated holes are available on the BiVO 4 electrode surface for HMF oxidation.

Photo-Reduction Reactions
It is interesting to find that most of studies focus on the oxidation of furan-based feedstocks, and few literatures have been reported in regard of photocatalytic reduction processes. In terms of thermodynamics and kinetics, the hydrogenation of C C group is much more favorable than that of C O double bond. Furthermore, the competitive adsorption of C C and C O groups at the catalyst surface also remains a challenge. Guo et al. first reported that photocatalytic reduction of HMF into DHMF can be achieved by using a platinum catalyst supported on graphitic carbon nitride (Pt/g-C 3 N 4 ) with trimethylamine as a sacrificial electron donor . The Pt/g-C 3 N 4 , as tandem catalyst, not only promoted the photo-induced water splitting to produce hydrogen but also facilitated the HMF reduction by activating the produced hydrogen. However, the catalytic performance of Pt/g-C 3 N 4 was not ideal, and the yield of DHMF was low (6.5% yield with TOF of 0.457 h −1 at 80°C). Thus, there is substantial interest in designing a more effective photocatalyst for the reduction of HMF.
In the heterogeneous photocatalytic furan derivatives reduction reaction, there are two main photo-reduction mechanisms. The first is the direct hydrogen transfer from the hydrogen donor to the acceptor via a formation of a six-membered transition state. For example, Zhang and co-workers reported that mesoporous carbon supported metallic Cu and Cu 2 O NPs have a synergistic effect on the selective hydrogenation of furfural (FAL) to furfuryl alcohol (FOL) under solar light irradiation with isopropanol as hydrogen source (Zhang and Li, 2019). Cu act as a bridge to promote the electron transfer from Cu 2 O to carbon support due to the work function difference between Cu and Cu 2 O, and the visible light irradiation facilitates this process. Meanwhile, the electron-rich mesoporous carbon favors the adsorption of C O bond in the terminal of FAL. The separated photogenerated holes oxidize the isopropanol to yield acetone and protons, which then transfer to the mesoporous carbon adsorbed C O bond to form FOL through a six-membered transition state (Figure 11). After 14 h reaction, 94.3% of FAL was consumed and a FOL yield of 90.9% was achieved.
Alternatively, the hydrogen from external hydrogen donor could be oxidized to form activated hydrogen species on catalyst surface, and then participate in the reduction reaction, namely indirect hydrogen transfer. Wang et al. reported that a Cu-H species can be formed on the surface of carbon-coated Cu NPs insitu by activating H 2 with hot electrons generated by LSPR effect. The C O group of FAL adsorbed on Cu is then attacked by two H atoms to produce FOL ( Figure 12). Decreased yield of FOL was observed when adding a hydrogen abstracting agent 2,2′,6,6-FIGURE 9 | Proposed mechanism for the photo-oxidation of HMF into DFF promoted by MnO 2 nanorods. Higher pyrolysis temperature led to thicker carbon cladding, hindering light absorption, and a lower pyrolysis temperature of 400°C favored the generation of Cu (Ⅱ) species, which showed weaker performance in the hydrogenation of FAL than the 600°C-formed metallic Cu. DFT results confirmed that the strong s-p-d hybridization facilitate the electron transfer between exposed Cu atoms and the C support and enhance the stability of Cu NPs . A SiC-supported Au NPs (Au/SiC) exhibited high selectivity (>90%) for the photocatalytic hydrogenation of a series of α,β-unsaturated aldehydes such as HMF and FAL under visible light irradiation (Hao et al., 2016). The elaborate designed plasmonic metal/semiconductor structure forms a built-in electric field at the Au/SiC interface, which enhances the directional electron transfer. LSPR leads to the injection of hot electrons into the CB of SiC, resulting in the formation of positively charged Au NPs and electron-rich active sites at the interface of Au/SiC. The high electron density and steric hindrance effect at the Au/SiC interface are beneficial for the adsorption of terminal C O groups rather than the bulky rings. Isopropanol can be oxidized on the positively charged Au surface to generate acetone and reactive hydrogen species. The latter is available for the reduction of aldehyde group to obtain FOL. A similar mechanism was also demonstrated in an Au/CuCo 2 O 4 system (Hu et al., 2020), where a nearly 100% selectivity to DHMF and a 93% conversion to HMF were achieved within 1 h.

CONCLUSIONS AND PERSPECTIVES
Carbohydrate, including cellulose and hemicellulose, constitute more than 60% of lignocellulosic biomass and represent the most abundant source of renewable carbon in nature. Abundant and free solar light as the driving force, unique photogenerated reactive species (e.g., h + , e − , • OH, • O 2 − , and 1 O 2 ) and mild reaction conditions make photocatalysis a "dream reaction". Using solar energy to selectively convert lignocellulosic biomass into value-added chemicals and fuels via photocatalysis at room temperature and pressure is an innovative but proven idea that could bring some merits from an energy and environmental perspective, such as reducing our carbon footprint and dependence on fossil fuels. Although the production of fine chemicals and biofuels is still small and far from satisfying the needs of society, it can be safely expected that biofuels will become the most important part of the global energy market in the foreseeable future. Despite tremendous efforts have been devoted in this field, photocatalytic transformation of biomass is still at an early stage and lags behind in terms of efficiency and cost. What is most surprising is that, to the best of our knowledge, few studies involving photocatalytic hemicellulose have been reported. There is still a long and tough way to go to achieve industrial applications.
(1) TiO 2 and g-C 3 N 4 based materials are the most frequently used photocatalysts in biomass valorization because of their excellent performance in both oxidative and reductive reactions. But it is not difficult to find that the lack of innovation in catalyst design in recent years has presented a dilemma of old wine in new bottles. Thus, it is extremely important to design novel, effective and noble-metal-free photocatalysts to convert carbohydrates to valuable products with high conversion and selectivity under the guidance of pairing the band edge positions of semiconductors with redox levels of the targeted oxidation and reduction half reactions. Meanwhile, achieving one-pot reaction to avoid costly intermediate separation processes is also worth consideration.
(2) In the future, advanced characterization techniques such as SEM, TEM, XRD, XPS, FT-IR and DRIFT can be deliberately combined with quantum chemical calculations, i.e., density functional theory, to get an atomic-level insight into the structure-property relationships of photocatalysts, favorable reaction pathways and the interactions between reactants and catalysts (Lu et al., 2016;Liu et al., 2018). Moreover, the application of accurate detection technologies (i.e., in situ  NMR spectroscopy, electron paramagnetic resonance (EPR), transient absorption spectroscopy (TAS)) to trace generated intermediates and active surface-bound radicals will help to elucidate the reaction mechanism of biomass photocatalysis (Xia et al., 2020). Subsequent product separation, purification and identification should also be followed up (GC-MS, HPLC, GPC, 2D HSQC NMR). (3) The engineering design and scale-up of the reactor for the conversion of biomass still need to be modified due to some foremost issues such as phase interaction between biomass and catalyst, heat and mass transfer. For traditional batch reactor, higher transparency for the irradiated light, lower light scattering and easy operation should be the basic requirements. On the other hand, due to the excellent control over photon and heat and mass transfer, designing microfluidic reactors holds great potential in improving the performance of photocatalytic system (Colmenares et al., 2017). Currently, studies supported by microfluidic reactors are still rare (Nguyen et al., 2014;Li et al., 2020), but the development of microfluidic reactors to improve the photocatalytic conversion of biomass is an important research direction in the near future. (4) The selection of an appropriate reaction temperature range always plays an important role in product selectivity; generally speaking, too high a temperature will lead to the generation of undesired byproducts, while too low a temperature will prolong the reaction time and greatly reduce the catalyst activity. Reaction time optimization is also extremely important since it determines the contact time between the reactants and the photocatalyst, which is likely to alter the product distribution. (5) Solvent is essential for most organic synthesis and catalytic reactions. It plays the role of facilitating heat and mass transfer and allow the reactant molecules to be homogeneously dispersed in the reaction medium. A good solvent for photocatalytic carbohydrates transformation should follow several requirements: 1) good solubility for carbohydrates and its derivatives; 2) be compatible with the chosen light source and reactor (solvent is neither a strong light absorber, nor does it corrode the reactor wall); 3) be physically and chemically stable under photocatalytic conditions; 4) nontoxic, low-cost, can be easily recycled and reused (Su et al., 2014). Water is the most abundant and environmental-friendly solvent, but receive less attention due to weaker dissolving ability than some organic solvents (ethanol, acetonitrile, trifluorotoluene) and the potential for the generation of nonselective hydroxyl radicals. (6) The conversion and selectivity of photocatalytic biomass transformation are not yet as high as those of conventional thermocatalysis. If catalytic production of fuels and chemicals from biomass can be scale up from laboratory to large biorefineries, in-depth studies of mixing efficiency, reaction mechanisms, reaction thermodynamics and kinetics, heat and mass transport phenomena in large reactors are required (Su et al., 2014).
In this review, recent advances on the photocatalytic production of important organic acids and furan chemicals from carbohydrates and their derivatives have been well summarized. Although most of current transformation methods are theoretically and technically feasible, they are not economical. However, these contributive studies are now bringing us closer to a more promising future which is featured with biomass-based technologies.