Cobalt–Graphene Catalyst for Selective Hydrodeoxygenation of Guaiacol to Cyclohexanol

Herein, cobalt-reduced graphene oxide (rGO) catalyst was synthesized with a practical impregnation–calcination approach for the selective hydrodeoxygenation (HDO) of guaiacol to cyclohexanol. The synthesized Co/rGO was characterized by transmission electron microscopy (TEM), high-angle annular dark-field scanning TEM (HAADF-STEM), X-ray photoelectron spectroscopy (XPS), Raman spectroscopy, X-ray diffraction (XRD), and H2 temperature-programmed reduction (H2-TPR) analysis. According to the comprehensive characterization results, the catalyst contains single Co atoms in the graphene matrix and Co oxide nanoparticles (CoOx) on the graphene surface. The isolated Co atoms embedded in the rGO matrix form stable metal carbides (CoCx), which constitute catalytically active sites for hydrogenation. The rGO material with proper amounts of N heteroatoms and lattice defects becomes a suitable graphene material for fabricating the catalyst. The Co/rGO catalyst without prereduction treatment leads to the complete conversion of guaiacol with 93.2% selectivity to cyclohexanol under mild conditions. The remarkable HDO capability of the Co/rGO catalyst is attributed to the unique metal–acid synergy between the CoCx sites and the acid sites of the CoOx nanoparticles. The CoCx sites provide H while the acid sites of CoOx nanoparticles bind the C-O group of reactants to the surface, allowing easier C-O scission. The reaction pathways were characterized based on the observed reaction–product distributions. The effects of the process parameters on catalyst preparation and the HDO reaction, as well as the reusability of the catalyst, were systematically investigated.


Introduction
The excessive consumption of fossil resources results in diminishing petroleum supply and severe environmental pollution. As a renewable energy resource, abundant and carbon-neutral biomass has been extensively explored for the production of highly valuable chemicals and biofuels [1]. However, the production of biomass-derived compounds with good selectivity remains challenging because of the complex structures and diverse oxygenic groups in biomass-derived feedstock [2]. Currently, catalytic hydrodeoxygenation (HDO) is considered the most efficient approach for upgrading biomass derivatives, and the development of cost-effective catalysts is key to this process [3]. As a main component of biomass, lignin is a planted polymer composed of phenylpropanoid building units that potentially provide renewable six-ring compounds [4]. Given the complex structure of lignin, guaiacol (GUA; 2-methoxyphenol), which contains two common oxygenate groups in lignin: methoxy (C aryl -OCH 3 ) and phenolic (C aryl -OH) groups, is extensively used as a lignin model compound in catalytic studies of lignin derivatives [5].
HDO of GUA involves combinations of different reactions such as hydrogenation, hydrogenolysis, and dehydration. Different elementary reactions usually occur at different catalytic sites. Typically, the heterogeneous catalysts designed for HDO contain two functions: one is for hydrogen dissociation, while the other is for the C-O activation [6]. To date,

Catalytic HDO Tests
The HDO of the GUA reactions was conducted in an autocontrol reactor (50 mL; Beijing Century Senlang Experimental Apparatus Co. Ltd., Beijing, China). In a typical run, the reactor was loaded with guaiacol (300 mg), the catalyst (30 mg; without prereduction treatment) and n-dodecane (10 mL), and then pressurized with H 2 to 1 MPa at room temperature. The reactor was heated to the target temperature and allowed to operate for a specific time. The liquid products were mixed and diluted by ethanol, and then analyzed using an Agilent GC7820 gas chromatograph equipped with a flame ionization detector and SE-30 capillary column (Dalian Zhonghuida Scientific Instrument Co., Ltd., Dalian, China). n-Tetradecane was used as the internal standard. Recycle test of the catalyst was carried out as follows: after the reaction, the catalyst was recovered from the reaction mixture by centrifugal separation, washed with ethanol, and vacuum-dried without calcination for the subsequent run. The results were quantified as GUA conversion, product selectivity, and yield in molar percentage, based on the number of C6 rings in the substrate and products. The overall carbon balance of the products was in the range of experimental error (±3%). The guaiacol conversion (X GUA ), the Product i selectivity (S i ), and Product i yield (Y i ) were calculated as following Equations The results indicate the co-existence of various Co oxides in the nanoparticles, which are termed as CoO x . Single Co atoms embedded in the graphene sheet were confirmed by HAADF-STEM images (Figure 1d,e). The area marked with a circle is an example of a single Co atom. Figure 1e shows the homogeneous distribution and high density of isolated Co atoms in the graphene nanosheet. The prepared catalyst was calcined at 500 • C in N 2 , and thus, the reserved Co single metal atoms were not physically adsorbed on the graphene sheet; by contrast, they built strong chemical bonding configurations with the rGO sheet. rGO is derived from GO through reduction treatments. After the removal of the oxygenic groups, the sp 3 -hybridized C atoms were terminated with H bonds, forming lattice defects on the graphene plane. Chemical bonding of the isolated metal atoms with pristine intact graphene is not easy due to the high chemical stability of the graphene's honeycomb structure [41]. As confirmed theoretically and experimentally [41,42], the introduction of defective sites in the graphene matrix offers multifarious bonding configurations to guarantee the structural stability of metal atoms. The high density and homogeneous distribution of single Co metal atoms in the rGO sheet ( Figure 1e) correspond to the high density and homogeneous distribution of lattice defects in the rGO structure. We named the Co single metal atoms in the graphene matrix as CoC x sites, which are similar to metal carbides. (3)

Electron Microscope Images
The morphology of the Co2.5/rGO catalyst was characterized by TEM, HRTEM, and HAADF-STEM ( Figure 1). The catalyst shows a homogenous dispersion of nanoparticles ( Figure 1a) with an average size of 6.7 nm in the TEM image (Figure 1c). The HRTEM image (Figure 1b) exhibits the typical crystalline morphology of the nanoparticles. The lattice fringes close to 0.213 and 0.244 nm are ascribed to the CoO (200; JCPDS card: 65-2902) and Co3O4 (311) (JCPDS card: 43-1003) planes, respectively. The lattice fringes of metallic Co were not observed in the HRTEM images. The results indicate the co-existence of various Co oxides in the nanoparticles, which are termed as CoOx. Single Co atoms embedded in the graphene sheet were confirmed by HAADF-STEM images (Figure 1d and e). The area marked with a circle is an example of a single Co atom. Figure 1e shows the homogeneous distribution and high density of isolated Co atoms in the graphene nanosheet. The prepared catalyst was calcined at 500 °C in N2, and thus, the reserved Co single metal atoms were not physically adsorbed on the graphene sheet; by contrast, they built strong chemical bonding configurations with the rGO sheet. rGO is derived from GO through reduction treatments. After the removal of the oxygenic groups, the sp 3 -hybridized C atoms were terminated with H bonds, forming lattice defects on the graphene plane. Chemical bonding of the isolated metal atoms with pristine intact graphene is not easy due to the high chemical stability of the graphene's honeycomb structure [41]. As confirmed theoretically and experimentally [41,42], the introduction of defective sites in the graphene matrix offers multifarious bonding configurations to guarantee the structural stability of metal atoms. The high density and homogeneous distribution of single Co metal atoms in the rGO sheet ( Figure 1e) correspond to the high density and homogeneous distribution of lattice defects in the rGO structure. We named the Co single metal atoms in the graphene matrix as CoCx sites, which are similar to metal carbides.

XPS
The XPS results in Figure 2 provide information on the surface elemental constituents and valence state of the Co 2.5 /rGO catalyst. Full-scan XPS survey spectrum ( Figure S1 in the Supplementary Materials) shows the predominant presence of C, O, N, and Co elements. The Co 2+ is characterized by the Co 2p 3/2 peak at 782.1 eV, Co 2p 1/2 peak at 798.0 eV, and corresponding shake-up resonances at approximately 787.9 and 803.6 eV [43]. The prominent peaks around 780.3 and 795.3 eV are assigned to the Co 2p 3/2 and Co 2p 1/2 peaks of the Co 3+ configuration with an energy difference of 15 eV [43]. The Co 2p peaks of the bulk cobalt carbide (Co 2 C) appear at 778.4 and 793.4 eV, which are at the binding energy in metallic Co [44]. In the Co 2.5 /rGO catalyst, the Co single-atoms doped in the graphene matrix of rGO formed a cobalt-carbide analog, but no peaks related to the bulk cobalt carbide were observed ( Figure 2). A similar phenomenon was reported in a previous study, which showed that the Co XPS of the graphene-supported single Co atoms have two main Co 2p peaks at 780.9 and 796.2 eV with the peak spacing of 15.4 eV and shake-up satellite peaks, indicating that the single Co atoms coordinate with oxidation states [45]. The isolated Co atoms doped in the graphene matrix mainly form multifarious bonding configurations with the surrounding C, and the N and O heteroatoms mediate the Co atoms in oxidation states. Thus, the binding energy of the single Co atoms is larger than that of bulk cobalt carbide. Therefore, the Co 2+ and Co 3+ peaks of Co 2.5 /rGO in Figure 2a correspond to isolated Co atoms (CoC x ) and CoO x nanoparticles.

XPS
The XPS results in Figure 2 provide information on the surface elemental constituents and valence state of the Co2.5/rGO catalyst. Full-scan XPS survey spectrum ( Figure S1 in the Supplementary Materials) shows the predominant presence of C, O, N, and Co elements. The Co 2+ is characterized by the Co 2p3/2 peak at 782.1 eV, Co 2p1/2 peak at 798.0 eV, and corresponding shake-up resonances at approximately 787.9 and 803.6 eV [43]. The prominent peaks around 780.3 and 795.3 eV are assigned to the Co 2p3/2 and Co 2p1/2 peaks of the Co 3+ configuration with an energy difference of 15 eV [43]. The Co 2p peaks of the bulk cobalt carbide (Co2C) appear at 778.4 and 793.4 eV, which are at the binding energy in metallic Co [44]. In the Co2.5/rGO catalyst, the Co single-atoms doped in the graphene matrix of rGO formed a cobalt-carbide analog, but no peaks related to the bulk cobalt carbide were observed ( Figure 2). A similar phenomenon was reported in a previous study, which showed that the Co XPS of the graphene-supported single Co atoms have two main Co 2p peaks at 780.9 and 796.2 eV with the peak spacing of 15.4 eV and shakeup satellite peaks, indicating that the single Co atoms coordinate with oxidation states [45]. The isolated Co atoms doped in the graphene matrix mainly form multifarious bonding configurations with the surrounding C, and the N and O heteroatoms mediate the Co atoms in oxidation states. Thus, the binding energy of the single Co atoms is larger than that of bulk cobalt carbide. Therefore, the Co 2+ and Co 3+ peaks of Co2.5/rGO in Figure 2a correspond to isolated Co atoms (CoCx) and CoOx nanoparticles. The C 1s spectrum of the Co2.5/rGO catalyst ( Figure 2b) splits into several components. The main peak at 284.5 eV corresponds to the sp 2 C in graphene [46], indicating that most of the C atoms form conjugated honeycomb lattices of graphene. The peak at approximately 285.3 eV is assigned to sp 3 C, which originates from the lattice defects and edges of the graphene sheets [46]. The peaks from 286 eV to 288 eV correspond to the C−N and C−O bonds [41,46]. The small peak approximately around 283.3 eV is attributed to the carbidic C 1s signal [44], which in turn is attributed to the C atoms bonded with isolated Co atoms of the Co2.5/rGO catalyst. Figure 3 compares the Raman spectra of graphite, rGO, and Co2.5/rGO. Highly ordered graphite has a G-band peak at approximately 1580 cm −1 , corresponding to the inphase vibration of the sp 2 carbon lattice, and a weak D-band peak at approximately 1350 cm −1 , corresponding to the sp 3 carbons caused by the defects on the graphite edges [47]. The rGO material contains a significant fraction of the sp 3 amorphous carbons mainly The C 1s spectrum of the Co 2.5 /rGO catalyst ( Figure 2b) splits into several components. The main peak at 284.5 eV corresponds to the sp 2 C in graphene [46], indicating that most of the C atoms form conjugated honeycomb lattices of graphene. The peak at approximately 285.3 eV is assigned to sp 3 C, which originates from the lattice defects and edges of the graphene sheets [46]. The peaks from 286 eV to 288 eV correspond to the C−N and C−O bonds [41,46]. The small peak approximately around 283.3 eV is attributed to the carbidic C 1s signal [44], which in turn is attributed to the C atoms bonded with isolated Co atoms of the Co 2.5 /rGO catalyst. Figure 3 compares the Raman spectra of graphite, rGO, and Co 2.5 /rGO. Highly ordered graphite has a G-band peak at approximately 1580 cm −1 , corresponding to the in-phase vibration of the sp 2 carbon lattice, and a weak D-band peak at approximately 1350 cm −1 , corresponding to the sp 3 carbons caused by the defects on the graphite edges [47]. The rGO material contains a significant fraction of the sp 3 amorphous carbons mainly generated by edges and lattice defects, and thus the rGO sample shows a noticeable D band. The ratio of the peak intensity of the D band to that of the G band (I D /I G ) is inversely proportional to the perfection of the graphene's honeycomb lattice [47,48]. As shown in Figure 3, the I D /I G value of the Co 2.5 /rGO catalyst is close to that of the rGO support, confirming that Nanomaterials 2022, 12, 3388 6 of 16 the isolated Co atoms do not attack the pristine sp 2 carbon lattice of graphene but are embedded in the defective positions as substitutional or interstitial dopants through the construction of "metal vacancy" heterostructures. Hence, rGO with appropriate lattice defects is an ideal substrate for anchoring isolated Co atoms. Nanomaterials 2022, 12, x 6 of 16 generated by edges and lattice defects, and thus the rGO sample shows a noticeable D band. The ratio of the peak intensity of the D band to that of the G band (ID/IG) is inversely proportional to the perfection of the graphene's honeycomb lattice [47,48]. As shown in Figure 3, the ID/IG value of the Co2.5/rGO catalyst is close to that of the rGO support, confirming that the isolated Co atoms do not attack the pristine sp 2 carbon lattice of graphene but are embedded in the defective positions as substitutional or interstitial dopants through the construction of "metal vacancy" heterostructures. Hence, rGO with appropriate lattice defects is an ideal substrate for anchoring isolated Co atoms.

XRD
The crystal structures of graphite, rGO, and Co2.5/rGO were investigated through XRD analysis ( Figure 4). Graphite has a sharp peak at a 2 theta of 26.4°, corresponding to the (002) plane of graphite (JCPDS PDF card: 41-1487). The XRD pattern of rGO shows trace amounts of graphite (002) diffraction signals with a widened peak and a low-degreeshift 2 theta, showing that rGO has a multilayer structure with increased lattice dimensions. The XRD peaks of the Co2.5/rGO sample show a slightly enhanced and high-degreeshift 2 theta of the (002) plane compared with that of rGO, which is caused by the stacking thickness of the graphene layers during the preparation of the catalyst. In addition to the graphitic diffraction, the XRD pattern of the Co2.5/rGO catalyst shows no diffraction signals corresponding to Co-related diffraction, indicating the small crystalline sizes and weak crystallinity of the Co species. This finding is in accordance with the TEM images ( Figure 1a).

XRD
The crystal structures of graphite, rGO, and Co 2.5 /rGO were investigated through XRD analysis ( Figure 4). Graphite has a sharp peak at a 2 theta of 26.4 • , corresponding to the (002) plane of graphite (JCPDS PDF card: 41-1487). The XRD pattern of rGO shows trace amounts of graphite (002) diffraction signals with a widened peak and a low-degree-shift 2 theta, showing that rGO has a multilayer structure with increased lattice dimensions. The XRD peaks of the Co 2.5 /rGO sample show a slightly enhanced and high-degree-shift 2 theta of the (002) plane compared with that of rGO, which is caused by the stacking thickness of the graphene layers during the preparation of the catalyst. In addition to the graphitic diffraction, the XRD pattern of the Co 2.5 /rGO catalyst shows no diffraction signals corresponding to Co-related diffraction, indicating the small crystalline sizes and weak crystallinity of the Co species. This finding is in accordance with the TEM images ( Figure 1a). generated by edges and lattice defects, and thus the rGO sample shows a noticeable D band. The ratio of the peak intensity of the D band to that of the G band (ID/IG) is inversely proportional to the perfection of the graphene's honeycomb lattice [47,48]. As shown in Figure 3, the ID/IG value of the Co2.5/rGO catalyst is close to that of the rGO support, confirming that the isolated Co atoms do not attack the pristine sp 2 carbon lattice of graphene but are embedded in the defective positions as substitutional or interstitial dopants through the construction of "metal vacancy" heterostructures. Hence, rGO with appropriate lattice defects is an ideal substrate for anchoring isolated Co atoms.

XRD
The crystal structures of graphite, rGO, and Co2.5/rGO were investigated through XRD analysis ( Figure 4). Graphite has a sharp peak at a 2 theta of 26.4°, corresponding to the (002) plane of graphite (JCPDS PDF card: 41-1487). The XRD pattern of rGO shows trace amounts of graphite (002) diffraction signals with a widened peak and a low-degreeshift 2 theta, showing that rGO has a multilayer structure with increased lattice dimensions. The XRD peaks of the Co2.5/rGO sample show a slightly enhanced and high-degreeshift 2 theta of the (002) plane compared with that of rGO, which is caused by the stacking thickness of the graphene layers during the preparation of the catalyst. In addition to the graphitic diffraction, the XRD pattern of the Co2.5/rGO catalyst shows no diffraction signals corresponding to Co-related diffraction, indicating the small crystalline sizes and weak crystallinity of the Co species. This finding is in accordance with the TEM images ( Figure 1a).

H 2 -TPR
The TPR patterns of the Co 2.5 /rGO catalyst are shown in Figure 5. rGO has certain oxygenic groups on its surface and periphery [49]. The hydrogenolysis of the -OH, -O-, and -COOH groups involves H 2 consumption starting at 400 • C. In the Co 2.5 /rGO catalyst, rGO and Co oxides consume H 2 . Moreover, according to the GC analysis, CH 4 exists in the TPR exhaust gas (Figure 5b). Thus, the H 2 consumption of the Co 2.5 /rGO sample includes the decomposition of oxygenic groups from rGO, reduction of Co oxides, and formation of CH 4 . The multipeak reduction of Co oxides corresponds to the step-step reduction of Co 2 O 3 /Co 3 O 4 to CoO and CoO to metallic Co [50]. A mixture of metallic Co powder (pre-reduced from CoO x powder) and rGO cannot generate CH 4 at a temperature of approximately 530 • C; therefore, the formation of CH 4 is a representative feature of the Co 2.5 /rGO sample during the H 2 -TPR test. We speculate that the single Co atoms embedded in the graphene plane catalyze the bonded C atoms of rGO to form CH 4.

H2-TPR
The TPR patterns of the Co2.5/rGO catalyst are shown in Figure 5. rGO has certain oxygenic groups on its surface and periphery [49]. The hydrogenolysis of the -OH, -O-, and -COOH groups involves H2 consumption starting at 400 °C. In the Co2.5/rGO catalyst, rGO and Co oxides consume H2. Moreover, according to the GC analysis, CH4 exists in the TPR exhaust gas (Figure 5 b). Thus, the H2 consumption of the Co2.5/rGO sample includes the decomposition of oxygenic groups from rGO, reduction of Co oxides, and formation of CH4. The multipeak reduction of Co oxides corresponds to the step-step reduction of Co2O3/Co3O4 to CoO and CoO to metallic Co [50]. A mixture of metallic Co powder (pre-reduced from CoOx powder) and rGO cannot generate CH4 at a temperature of approximately 530 °C; therefore, the formation of CH4 is a representative feature of the Co2.5/rGO sample during the H2-TPR test. We speculate that the single Co atoms embedded in the graphene plane catalyze the bonded C atoms of rGO to form CH4.

Elemental Analyses
The three graphene materials, rGO, rGO*, and Gr, have similar specific surface areas but different elemental compositions ( Table 1). The O atoms mainly come from the oxygenic groups, such as -OH and -COOH. The H atoms originate from the dangling H atoms for the termination of the dangling bonds of the graphene sheet edges, defects, and some groups. The rGO and rGO* materials have similar C, H, and O contents, but rGO contains 3.7 mol% N atoms. N was incorporated into the rGO network through the hydrothermal reduction of GO using hydrazine hydrate and ammonia as reducing reagents [51]. The comparison of rGO with Gr shows that rGO contains more H than Gr dose. More H indicates more lattice defects because the C dangling bonds are terminated by the H atoms.

Elemental Analyses
The three graphene materials, rGO, rGO*, and Gr, have similar specific surface areas but different elemental compositions ( Table 1). The O atoms mainly come from the oxygenic groups, such as -OH and -COOH. The H atoms originate from the dangling H atoms for the termination of the dangling bonds of the graphene sheet edges, defects, and some groups. The rGO and rGO* materials have similar C, H, and O contents, but rGO contains 3.7 mol% N atoms. N was incorporated into the rGO network through the hydrothermal reduction of GO using hydrazine hydrate and ammonia as reducing reagents [51]. The comparison of rGO with Gr shows that rGO contains more H than Gr dose. More H indicates more lattice defects because the C dangling bonds are terminated by the H atoms. GUA contains C aryl -OH, C aryl -OCH 3, and C aryl O-CH 3 bonds and an aromatic ring in its structure. The catalytic conversion of GUA may generate a wide range of products over metal catalysts, including those formed without oxygen removal, partial oxygen removal, and complete oxygen removal. These products are generated by different reaction pathways, such as hydrogenation (HYD), demethoxylation (DMO), and dehydration (DHY). The possible reaction pathways for the HDO of GUA were explored based on the product distributions. As shown in Scheme 1, the main pathway involves the cleavage of C aryl -Nanomaterials 2022, 12, 3388 8 of 16 OCH 3 bond to phenol (Ph) through HYD and DMO followed by the saturation of the aromatic ring to CYHAOL through HYD; the other involves the saturation of the aromatic ring to 2-methoxycyclohexanol (MOCYHOL) through HYD and then the cleavage of C alkyl -OCH 3 bond to CYHAOL. The DHY of CYHAOL generates a CYHA by-product.
GUA contains Caryl-OH, Caryl-OCH3, and Caryl O-CH3 bonds and an aromatic ring in its structure. The catalytic conversion of GUA may generate a wide range of products over metal catalysts, including those formed without oxygen removal, partial oxygen removal, and complete oxygen removal. These products are generated by different reaction pathways, such as hydrogenation (HYD), demethoxylation (DMO), and dehydration (DHY). The possible reaction pathways for the HDO of GUA were explored based on the product distributions. As shown in Scheme 1, the main pathway involves the cleavage of Caryl-OCH3 bond to phenol (Ph) through HYD and DMO followed by the saturation of the aromatic ring to CYHAOL through HYD; the other involves the saturation of the aromatic ring to 2-methoxycyclohexanol (MOCYHOL) through HYD and then the cleavage of Calkyl-OCH3 bond to CYHAOL. The DHY of CYHAOL generates a CYHA by-product.

Scheme 1.
Possible reaction pathways in the HDO of GUA over the Co/rGO catalyst.

Catalyst Screening Study
The catalysts' performance for HDO of GUA depends on two types of active sites: one is the metal-like site that readily activates hydrogen, while the other type of site nearby, such as an acid site, is responsible for the C-O bond activation [6,52]. For screening the functional catalyst, Fe, Co and Ni supported on rGO, and Co supported on different supports were tested in HDO of GUA. The catalytic performance of the prepared catalysts is presented in Table 2. Fe, Co, and Ni are eight-group elements with outer electronic structures of 3d 6 4s 2 , 3d 7 4s 2 , and 3d 8 4s 2 , respectively, and the three corresponding catalysts have similar specific surface areas but different catalytic activities (Table 1, entries 1-3). When the reaction occurs over the Co2.5/rGO catalyst at 200 ℃ and 1.0 MPa H2 pressure for 2 h, GUA is completely converted, and the yield of CYHAOL reaches 93.2 mol%. The Ni2.5/rGO catalyst achieves 90.1 mol% conversion with 62.4 mol% and 35.1 mol% selectivity to CYHAOL and MOCYHOL, respectively. Ni catalysts usually catalyze total HYD to MOCYHOL as a major side reaction, especially at lower reaction temperatures [26]. The Fe2.5/GO catalyst only converts 1.8 mol% of GUA. CYHA as a by-product is rarely detected in the reaction catalyzed by Co2.5/rGO; therefore, Co is the most suitable transition metal to be supported on rGO for HDO reaction.

Catalyst Screening Study
The catalysts' performance for HDO of GUA depends on two types of active sites: one is the metal-like site that readily activates hydrogen, while the other type of site nearby, such as an acid site, is responsible for the C-O bond activation [6,52]. For screening the functional catalyst, Fe, Co and Ni supported on rGO, and Co supported on different supports were tested in HDO of GUA. The catalytic performance of the prepared catalysts is presented in Table 2. Fe, Co, and Ni are eight-group elements with outer electronic structures of 3d 6 4s 2 , 3d 7 4s 2 , and 3d 8 4s 2 , respectively, and the three corresponding catalysts have similar specific surface areas but different catalytic activities (Table 1, entries 1-3). When the reaction occurs over the Co 2.5 /rGO catalyst at 200°C and 1.0 MPa H 2 pressure for 2 h, GUA is completely converted, and the yield of CYHAOL reaches 93.2 mol%. The Ni 2.5 /rGO catalyst achieves 90.1 mol% conversion with 62.4 mol% and 35.1 mol% selectivity to CYHAOL and MOCYHOL, respectively. Ni catalysts usually catalyze total HYD to MOCYHOL as a major side reaction, especially at lower reaction temperatures [26]. The Fe 2.5 /GO catalyst only converts 1.8 mol% of GUA. CYHA as a by-product is rarely detected in the reaction catalyzed by Co 2.5 /rGO; therefore, Co is the most suitable transition metal to be supported on rGO for HDO reaction. The support material is another decisive factor for determining catalyst activity. Compared with the Co 2.5 /rGO catalyst, the Co 2.5 /Al 2 O 3 and Co 2.5 /HY samples ( Table 2, entries 4 and 5) show much lower activities. In general, a pristine metal species is an active component in the hydrogenation reaction; thus, the catalysts are prereduced or originally contain metals before they are used in GUA hydrogenation [13][14][15][16][17][18][19][20][21][22][23][24][25][26]. In this study, all the catalysts were used without prereduction treatments. Co in the Co 2.5 /Al 2 O 3 and Co 2.5 /HY catalysts is in an oxide state, and thus, the catalysts exhibit extremely low activities because of insufficient metallic Co. However, the Co species in Co 2.5 /rGO contains no metallic Co according to the HRTEM, XPS, and XRD characterization results, and the catalyst was used without prereduction treatment. Co 2.5 /rGO was directly exposed to air during collection and weighing. Therefore, the active sites of Co 2.5 /rGO are different from the frequently used type of catalysts requiring reduction treatment. Transition metal carbides, which are prepared by incorporating carbon atoms into the lattices of transition metals, have been demonstrated as promising catalysts for biomass conversions, especially in the C-C, C-O-C, and C-O-H bonds cleavage reactions [53]. We speculate that the catalytic component of Co 2.5 /rGO for hydrogenation originates from a Co-carbide analog (CoC x ) formed by embedding Co single atoms in the graphene matrix. The poor reaction result over the physical CoO x + rGO mixture (Table 2, entry 6) confirms the critical role of the CoC x sites in hydrogenation catalysis. Thus, rGO as a support is involved in the construction of the CoC x sites, whereas Al 2 O 3 and HY support lack such functions.
The above results indicate that using rGO as a support is one of the decisive factors in determining the performance of the Co 2.5 /rGO catalyst. According to the elemental analysis results (Table 1), rGO * is an N-free material, and rGO contains 3.7 mol% N. Co 2.5 /rGO * (Table 2, entry 7) achieves 85.0 mol% conversion and 59.5 mol% CYHAOL yield, manifesting that the doped N atoms are not an indispensable factor in catalytic function. However, the improved activity of the Co 2.5 /rGO catalyst as compared with that of Co 2.5 /rGO * confirms a positive role of N doping in graphene. Transition metal nitrides have been explored in HDO conversions [28]. Moreover, the N atoms merged within the graphene matrix disrupt the electronic neutrality of adjacent carbon atoms [54,55], which manipulate the electronic status of CoC x and CoO x in higher electronic density. This effect improves the activity of the Co 2.5 /rGO catalyst. The O functional groups of rGO may improve the dispersion and stability of the supported CoO x nanoparticles through an anchoring effect during impregnation and calcination. In brief, rGO with appropriate amounts of N and O heteroatoms is a suitable graphene material as a support.
In order to confirm the hypothesis that the hydrogenation function of Co 2.5 /rGO is provided by CoC x , the Co 2.5 /Gr catalyst prepared with a commercial graphene (Gr) was investigated for clarity. The conversion and CYHAOL yield over Co 2.5 /Gr (Table 1, entry 8) is much less than that over Co 2.5 /rGO. The catalyst-specific surface area of Co 2.5 /Gr is 341 m 2 ·g -1 , and the average particle size of CoO x nanoparticles on Gr surface ( Figure S2a) is 6.5 nm, which indicates that the low activity of Co 2.5 /Gr is not caused by the small specific surface area and the low dispersion of CoO x components. Supposing the hydrogenation function of Co 2.5 /rGO derives from the in situ reduction of CoO x nanoparticles to metallic Co during reaction, then Co 2.5 /Gr should be as good-or as bad-as Co 2.5 /rGO; therefore, some critical factors determine the catalysts' performance. The biggest difference between rGO and Gr is that rGO can anchor more single-metal atoms than Gr does. The evidence is shown in the HAADF-STEM image of Co 2.5 /rGO. (Figure 1e) shows the high-density distribution of the isolated Co atoms in the graphene matrix, while the number of the isolated Co atoms decreases considerably in the image of Co 2.5 /Gr ( Figure S2b). The more Co single atoms in the graphene matrix, the more Co-carbide analog (CoC x ) the catalyst contains. Transition metal carbides have been compared to platinum group metals, showing similar catalytic properties, which means that they could be promising catalysts for HDO reactions [52]; therefore, Co 2.5 /Gr exhibits weak catalytic activity due to the low population of the CoC x sites.
The data in entry 6 of Table 2 show that CoO x fails to conduct the HDO of GUA without the participation of CoC x ; however, the CoO x nanoparticles on rGO surface may improve the ability of the Co 2.5 /rGO catalyst to break C aryal -OCH 3 bond. HDO was proposed to need a bifunctional catalyst where acid sites are required for C-O bond activation, allowing easier C-O scission [6]. CoO x contains acidity according to the NH 3 -TPD result ( Figure S3a) and can adsorb the oxygenic groups in a GUA molecule through acid-base-pairing interactions, facilitating the cleavage of the C-O bond. Therefore, CoO x, in combination with CoC x, plays a role in establishing the metal−acid bifunction of the catalyst. A similar bifunctional catalyst was established for HDO reaction based on molybdenum carbide and oxide [56]. When Ph and CYHAOL are used as feedstock and the reaction is run at 200 • C and 1 MPa H 2 for 2 h, Ph is completely converted to CYHAOL over the Co 2.5 /rGO catalyst, whereas CYHAOL is almost not converted. Cleavage of the C aryl -OH bond needs strong acidic components such as HZSM-5 zeolite [2,28], while the mild acid of CoO x appropriately avoids cleaving the C aryl -OH bond. In this reaction, especially carried out in milder conditions, the basicity of support can affect the selectivity of bond breaking [6,57,58]. The presence of base promotes the demethoxylation step and suppresses the unselective C-O dissociation [57]. The basicity of CoO x is not obvious according to the CO 2 -TPD result ( Figure S3b), but N doping incorporates basic sites in the graphene texture [26], which potentially favors the C aryl -OCH 3 bond cleavage, making the Co 2.5 /rGO catalyst has good selectivity to CYHAOL. Scheme 2 shows the main catalysis in the selective HDO of GUA to CYHAOL over the Co 2.5 /rGO catalyst, in which Ph is used as the intermediate. H 2 dissociates to active H species on the CoC x sites; the C aryl -OCH 3 bond of GUA is activated by the oxophilic acid sites of CoO x and dissociated by active H. The metal-acid bifunction of the Co 2.5 /rGO catalyst in the HDO of GUA to CYHAOL is attributed to the unique synergy between the CoC x sites and the acid sites of the CoO x nanoparticles. The Co 2.5 /rGO catalyst selectively saturates the aromatic ring and dominantly cleaves C aryl -OCH 3 bond in the GUA. These processes result in a good performance in CYHAOL production.
the isolated Co atoms decreases considerably in the image of Co2.5/Gr ( Figure S2b). The more Co single atoms in the graphene matrix, the more Co-carbide analog (CoCx) the catalyst contains. Transition metal carbides have been compared to platinum group metals, showing similar catalytic properties, which means that they could be promising catalysts for HDO reactions [52]; therefore, Co2.5/Gr exhibits weak catalytic activity due to the low population of the CoCx sites.
The data in entry 6 of Table 2 show that CoOx fails to conduct the HDO of GUA without the participation of CoCx; however, the CoOx nanoparticles on rGO surface may improve the ability of the Co2.5/rGO catalyst to break Caryal-OCH3 bond. HDO was proposed to need a bifunctional catalyst where acid sites are required for C-O bond activation, allowing easier C-O scission [6]. CoOx contains acidity according to the NH3-TPD result ( Figure S3a) and can adsorb the oxygenic groups in a GUA molecule through acidbase-pairing interactions, facilitating the cleavage of the C-O bond. Therefore, CoOx, in combination with CoCx, plays a role in establishing the metal−acid bifunction of the catalyst. A similar bifunctional catalyst was established for HDO reaction based on molybdenum carbide and oxide [56]. When Ph and CYHAOL are used as feedstock and the reaction is run at 200 °C and 1 MPa H2 for 2 h, Ph is completely converted to CYHAOL over the Co2.5/rGO catalyst, whereas CYHAOL is almost not converted. Cleavage of the Caryl-OH bond needs strong acidic components such as HZSM-5 zeolite [2,28], while the mild acid of CoOx appropriately avoids cleaving the Caryl-OH bond. In this reaction, especially carried out in milder conditions, the basicity of support can affect the selectivity of bond breaking [6,57,58]. The presence of base promotes the demethoxylation step and suppresses the unselective C-O dissociation [57]. The basicity of CoOx is not obvious according to the CO2-TPD result ( Figure S3b), but N doping incorporates basic sites in the graphene texture [26], which potentially favors the Caryl-OCH3 bond cleavage, making the Co2.5/rGO catalyst has good selectivity to CYHAOL. Scheme 2 shows the main catalysis in the selective HDO of GUA to CYHAOL over the Co2.5/rGO catalyst, in which Ph is used as the intermediate. H2 dissociates to active H species on the CoCx sites; the Caryl-OCH3 bond of GUA is activated by the oxophilic acid sites of CoOx and dissociated by active H. The metal-acid bifunction of the Co2.5/rGO catalyst in the HDO of GUA to CYHAOL is attributed to the unique synergy between the CoCx sites and the acid sites of the CoOx nanoparticles. The Co2.5/rGO catalyst selectively saturates the aromatic ring and dominantly cleaves Caryl-OCH3 bond in the GUA. These processes result in a good performance in CYHAOL production.

Scheme 2.
Possible reaction mechanism of HDO of GUA over the Co2.5/rGO catalyst with Ph as the intermediate.

Optimization of Catalyst Preparation Conditions
The effects of the Co loading and calcination temperature on the Co/rGO catalysts were studied. The results are shown in Figure 6. Co/rGO catalysts with different Co loadings were prepared for the HDO of GUA (Figure 6a). When the Co loading increases from 1.0 mmol/g to 2.5 mmol/g, the yield of CYHAOL constantly increases and reaches 93.2 Scheme 2. Possible reaction mechanism of HDO of GUA over the Co 2.5 /rGO catalyst with Ph as the intermediate.

Optimization of Catalyst Preparation Conditions
The effects of the Co loading and calcination temperature on the Co/rGO catalysts were studied. The results are shown in Figure 6. Co/rGO catalysts with different Co loadings were prepared for the HDO of GUA (Figure 6a). When the Co loading increases from 1.0 mmol/g to 2.5 mmol/g, the yield of CYHAOL constantly increases and reaches 93.2 mol% at 2.5 mmol/g; further increase in Co loading has little contribution to the CYHAOL yield. Figure 6b shows the results obtained for the Co 2.5 /rGO catalysts at different calcination temperatures in N 2 . The conversion of GUA and the yield of CYHAOL shows volcano-like shapes when the calcination temperature is increased from 400 • C to 700 • C, achieving maximum values at 500 • C. The Co 2.5 /rGO catalyst calcinated at 400 • C in N 2 only leads to 4.6 mol% conversion of GUA. When the calcination temperature is above 500 • C, the aggregation of nanoparticles becomes obvious, as evidenced by the low specific surface areas of the samples calcined at 600 • C and 700 • C. The highest efficiency for the HDO of GUA was obtained using a catalyst with a loading of 2.5 mmol Co per gram of rGO at a calcination temperature of 500 • C.
achieving maximum values at 500 °C. The Co2.5/rGO catalyst calcinated at 400 °C in N2 only leads to 4.6 mol% conversion of GUA. When the calcination temperature is above 500 °C, the aggregation of nanoparticles becomes obvious, as evidenced by the low specific surface areas of the samples calcined at 600 °C and 700 °C. The highest efficiency for the HDO of GUA was obtained using a catalyst with a loading of 2.5 mmol Co per gram of rGO at a calcination temperature of 500 °C. Blanco et al. [26] prepared the Co/GOr and Co/GOr-N catalysts by wet impregnation of Co over reduced graphene oxide undoped and doped with N, calcination in N2 at 350 °C and reduction under H2 at 300 °C. As evidenced by the XRD pattern and CO chemisorption of the catalysts, the hydrogenation sites of Co/rGOr and Co/rGOr-N are metallic Co (Co 0 ) components. In our study, the Co2.5/rGO catalyst calcinated at 500 °C in N2 and used without H2-reduction pretreatment exhibits good activity. Metallic oxides or nitrates on carbonaceous supports may form metals by carbothermal reduction during calcination in inert gases. For example, NiFe alloy nanoparticles were prepared by calcination of a cellulose filter paper impregnated with Fe and Ni nitrates at 800 °C for 2 h under N2 [59]. The Co2.5/rGO catalyst calcinated at 500 °C in N2 does not contain metallic Co according to the XRD pattern and the lattice fringes in the HRTEM image of the catalyst. Therefore, the CoCx sites of Co2.5/rGO serve as the activity sites. The low activity of the Co2.5/rGO catalyst calcined at 400 °C in N2 manifests that the calcination temperature at 400 °C or below cannot provide sufficient energy for the formation of the CoCx sites, so the Co/GOr and Co/GOr-N catalysts calcinated at only 350 °C [26] can hardly contain the CoCx sites. Figure 7 shows the variations in GUA conversion and product distribution among different reaction conditions. As reaction time proceeds, GUA conversion rate increases rapidly and reaches 100 mol% at 2 h. Selectivity to the Ph intermediate product gradually decreases over time after the initial accumulation, accompanied by an increase in selectivity to CYHAOL. This result shows that Ph is an intermediate progressing toward CYHAOL. An increase in reaction temperature from 160 °C to 200 °C increases the conversion rate of GUA and yield of CYHAOL. When the reaction approaches the 100 mol% conversion, the improved reaction temperature slightly decreases the CYHAOL yield, indicating that CYHAOL is relatively stable over the temperature range from 200 °C to 240 °C. With increasing H2 pressure from 0.25 MPa to 4.0 MPa, the CYHAOL yield reaches 93.2 mol% at 1.0 MPa and then decreases gradually with pressure, and selectivity to Blanco et al. [26] prepared the Co/GOr and Co/GOr-N catalysts by wet impregnation of Co over reduced graphene oxide undoped and doped with N, calcination in N 2 at 350 • C and reduction under H 2 at 300 • C. As evidenced by the XRD pattern and CO chemisorption of the catalysts, the hydrogenation sites of Co/rGOr and Co/rGOr-N are metallic Co (Co 0 ) components. In our study, the Co 2.5 /rGO catalyst calcinated at 500 • C in N 2 and used without H 2 -reduction pretreatment exhibits good activity. Metallic oxides or nitrates on carbonaceous supports may form metals by carbothermal reduction during calcination in inert gases. For example, NiFe alloy nanoparticles were prepared by calcination of a cellulose filter paper impregnated with Fe and Ni nitrates at 800 • C for 2 h under N 2 [59]. The Co 2.5 /rGO catalyst calcinated at 500 • C in N 2 does not contain metallic Co according to the XRD pattern and the lattice fringes in the HRTEM image of the catalyst. Therefore, the CoC x sites of Co 2.5 /rGO serve as the activity sites. The low activity of the Co 2.5 /rGO catalyst calcined at 400 • C in N 2 manifests that the calcination temperature at 400 • C or below cannot provide sufficient energy for the formation of the CoC x sites, so the Co/GOr and Co/GOr-N catalysts calcinated at only 350 • C [26] can hardly contain the CoC x sites. Figure 7 shows the variations in GUA conversion and product distribution among different reaction conditions. As reaction time proceeds, GUA conversion rate increases rapidly and reaches 100 mol% at 2 h. Selectivity to the Ph intermediate product gradually decreases over time after the initial accumulation, accompanied by an increase in selectivity to CYHAOL. This result shows that Ph is an intermediate progressing toward CYHAOL. An increase in reaction temperature from 160 • C to 200 • C increases the conversion rate of GUA and yield of CYHAOL. When the reaction approaches the 100 mol% conversion, the improved reaction temperature slightly decreases the CYHAOL yield, indicating that CYHAOL is relatively stable over the temperature range from 200 • C to 240 • C. With increasing H 2 pressure from 0.25 MPa to 4.0 MPa, the CYHAOL yield reaches 93.2 mol% at 1.0 MPa and then decreases gradually with pressure, and selectivity to MOCYHOL increases. These effects indicate that high H 2 pressure intensifies the HYD saturation of the aromatic ring to form MOCYHOL. Theoretically, MOCYHOL should proceed toward CYHAOL through the cleavage of C alkyl -OCH 3 bond. However, the fully hydrogenated MOCYHOL is more stable than Ph, and the higher steric hindrance of MOCYHOL restrains the cleavage of C alkyl -OCH 3 bond. Therefore, the GUA-Ph-CYHAOL pathway is much more conducive to producing MOCYHOL than the GUA-MOCYHOL-CYHAOL pathway, and a lower H 2 pressure facilitates CYHAOL formation. saturation of the aromatic ring to form MOCYHOL. Theoretically, MOCYHOL should proceed toward CYHAOL through the cleavage of Calkyl-OCH3 bond. However, the fully hydrogenated MOCYHOL is more stable than Ph, and the higher steric hindrance of MOCYHOL restrains the cleavage of Calkyl-OCH3 bond. Therefore, the GUA-Ph-CYHAOL pathway is much more conducive to producing MOCYHOL than the GUA-MOCYHOL-CYHAOL pathway, and a lower H2 pressure facilitates CYHAOL formation.

Catalyst Recycle Study
Catalyst stability is a critical factor in developing conversion processes industrially. Sintering and leaching of metal components are major problems for liquid-phase HDO reactions [60]. Further, coking on the catalyst and structural degradation leads to the deactivation of the catalysts [60]. Figure 8 summarizes the characterization results of the spent Co2.5/rGO catalyst. The TEM image of the spent Co2.5/rGO catalyst shows the homogenous dispersion of nanoparticles ( Figure 8a) with a slightly larger average size of 7.6 nm (Figure 8b) as compared to that of a fresh one (6.7 nm). Fresh Co2.5/rGO catalyst contains no metallic Co (Figure 2a) and is used without prereduction treatment. A hypothesis states that Co2.5/rGO catalysts undergo in situ reduction to form metallic Co. However, the XPS spectra of the spent catalyst (Figure 8c) show no metallic Co, thereby confirming that the active sites for activating H2 is CoCx rather than metallic Co. Figure 8d shows the TG curves of the fresh and spent Co2.5/rGO catalysts. The weight loss rates mainly relate to the burning of rGO at approximately 310 °C, and the coke, if present, cannot be identified because it burnt with rGO together. However, a careful comparison of the two curves found that the weight loss of the spent catalyst is about 3% higher than that of the fresh one, which is caused by coke deposition on the spent catalyst. Figure 8e,f show the high density of isolated Co atoms in the graphene nanosheet of the spent catalyst.

Catalyst Recycle Study
Catalyst stability is a critical factor in developing conversion processes industrially. Sintering and leaching of metal components are major problems for liquid-phase HDO reactions [60]. Further, coking on the catalyst and structural degradation leads to the deactivation of the catalysts [60]. Figure 8 summarizes the characterization results of the spent Co 2.5 /rGO catalyst. The TEM image of the spent Co 2.5 /rGO catalyst shows the homogenous dispersion of nanoparticles ( Figure 8a) with a slightly larger average size of 7.6 nm (Figure 8b) as compared to that of a fresh one (6.7 nm). Fresh Co 2.5 /rGO catalyst contains no metallic Co (Figure 2a) and is used without prereduction treatment. A hypothesis states that Co 2.5 /rGO catalysts undergo in situ reduction to form metallic Co. However, the XPS spectra of the spent catalyst (Figure 8c) show no metallic Co, thereby confirming that the active sites for activating H 2 is CoC x rather than metallic Co. Figure 8d shows the TG curves of the fresh and spent Co 2.5 /rGO catalysts. The weight loss rates mainly relate to the burning of rGO at approximately 310 • C, and the coke, if present, cannot be identified because it burnt with rGO together. However, a careful comparison of the two curves found that the weight loss of the spent catalyst is about 3% higher than that of the fresh one, which is caused by coke deposition on the spent catalyst. Figure 8e,f show the high density of isolated Co atoms in the graphene nanosheet of the spent catalyst.
The reusability of the Co 2.5 /rGO catalyst was investigated during the HDO of the GUA (Figure 9). The catalyst recycling experiment was performed without regeneration. The conversion rate of GUA drops to 70.4 mol% in the repeated test (cycle 2), indicating the deactivation of the Co 2.5 /rGO catalyst. One concern regarding the causes of deactivation is that single Co atoms in the graphene matrix may move and aggregate during the thermodynamic process of a reaction, which leads to the loss of CoC x sites. A model reaction was performed to exclude this possibility. The fresh Co 2.5 /rGO catalyst was first processed under the following conditions: 10 mL of n-dodecane, 30 mg of the catalyst, 200 • C, and 1 MPa H 2 for 2 h. Then, 300 mg of GUA was added to the reactor, and the reaction proceeded under the same conditions. The result of "Pretreated" in Figure 9 shows that the catalyst maintains its activity, indicating that the single Co atoms embedded in the rGO lattice are stable under the reaction conditions, in accordance with the image of HAADF-STEM (Figure 8f). Thereby, coking is the main reason for the deactivation of the catalyst. Coking was also presented in previous studies on the catalytic HDO of GUA [21,[60][61][62][63].
that the active sites for activating H2 is CoCx rather than metallic Co. Figure 8d shows the TG curves of the fresh and spent Co2.5/rGO catalysts. The weight loss rates mainly relate to the burning of rGO at approximately 310 °C, and the coke, if present, cannot be identified because it burnt with rGO together. However, a careful comparison of the two curves found that the weight loss of the spent catalyst is about 3% higher than that of the fresh one, which is caused by coke deposition on the spent catalyst. Figure 8e,f show the high density of isolated Co atoms in the graphene nanosheet of the spent catalyst. The reusability of the Co2.5/rGO catalyst was investigated during the HDO of the GUA (Figure 9). The catalyst recycling experiment was performed without regeneration. The conversion rate of GUA drops to 70.4 mol% in the repeated test (cycle 2), indicating the deactivation of the Co2.5/rGO catalyst. One concern regarding the causes of deactivation is that single Co atoms in the graphene matrix may move and aggregate during the thermodynamic process of a reaction, which leads to the loss of CoCx sites. A model reaction was performed to exclude this possibility. The fresh Co2.5/rGO catalyst was first processed under the following conditions: 10 mL of n-dodecane, 30 mg of the catalyst, 200 °C, and 1 MPa H2 for 2 h. Then, 300 mg of GUA was added to the reactor, and the reaction proceeded under the same conditions. The result of "Pretreated" in Figure 9 shows that the catalyst maintains its activity, indicating that the single Co atoms embedded in the rGO lattice are stable under the reaction conditions, in accordance with the image of HAADF-STEM (Figure 8f). Thereby, coking is the main reason for the deactivation of the catalyst. Coking was also presented in previous studies on the catalytic HDO of GUA [21,[60][61][62][63].  The reusability of the Co2.5/rGO catalyst was investigated during the HDO of the GUA (Figure 9). The catalyst recycling experiment was performed without regeneration. The conversion rate of GUA drops to 70.4 mol% in the repeated test (cycle 2), indicating the deactivation of the Co2.5/rGO catalyst. One concern regarding the causes of deactivation is that single Co atoms in the graphene matrix may move and aggregate during the thermodynamic process of a reaction, which leads to the loss of CoCx sites. A model reaction was performed to exclude this possibility. The fresh Co2.5/rGO catalyst was first processed under the following conditions: 10 mL of n-dodecane, 30 mg of the catalyst, 200 °C, and 1 MPa H2 for 2 h. Then, 300 mg of GUA was added to the reactor, and the reaction proceeded under the same conditions. The result of "Pretreated" in Figure 9 shows that the catalyst maintains its activity, indicating that the single Co atoms embedded in the rGO lattice are stable under the reaction conditions, in accordance with the image of HAADF-STEM (Figure 8f). Thereby, coking is the main reason for the deactivation of the catalyst. Coking was also presented in previous studies on the catalytic HDO of GUA [21,[60][61][62][63].

Conclusions
The catalytic HDO of GUA, a phenolic model compound of biomass lignin pyrolysis products, was investigated over Co/rGO catalysts synthesized through a practical impregnation-calcination approach. A series of characterization results revealed that the prepared Co/rGO catalyst contains single Co atoms embedded in the graphene matrix and CoOx nanoparticles on the graphene surface. The isolated Co atoms formed stable metal carbide analogs (CoCx) with rGO when the catalyst was calcined at 500 °C in N2. The

Conclusions
The catalytic HDO of GUA, a phenolic model compound of biomass lignin pyrolysis products, was investigated over Co/rGO catalysts synthesized through a practical impregnation-calcination approach. A series of characterization results revealed that the prepared Co/rGO catalyst contains single Co atoms embedded in the graphene matrix and CoO x nanoparticles on the graphene surface. The isolated Co atoms formed stable metal carbide analogs (CoC x ) with rGO when the catalyst was calcined at 500 • C in N 2 . The graphene texture is a decisive factor in determining the catalyst performance. rGO with proper N heteroatoms and lattice defects is a suitable material for catalyst fabrication. The optimized Co/rGO catalyst without prereduction treatment led to the complete conversion of GUA with 93.2% yield to CYHAOL under mild reaction conditions (200 • C and 1.0 MPa H 2 pressure for 2 h). The catalytic HDO of GUA to GYHAOL mainly occurred through the pathway using phenol as the intermediate. The Co/rGO catalyst possesses metal-acid bifunctional characteristics, by which the CoC x sites readily activate H 2 and provide active H, and the CoO x nanoparticles provide acid sites that are required to activate the C-O bond, allowing easier C-O scission by active H. The isolated Co atoms of the Co/rGO catalyst were stable under the reaction conditions of 200 • C and 1.0 MPa H 2 , which emphasized the feasibility of constructing the highly dispersed metal-carbide species through embedding single metal atoms in the graphene matrix.