Metal oxide nanoparticles embedded in porous carbon for sulfur absorption under hydrothermal conditions

MOx (M = Zn, Cu, Mn, Fe, Ce) nanoparticles (NPs) embedded in porous C with uniform diameter and dispersion were synthesized, with potential application as S-absorbents to protect catalysts from S-poisoning in catalytic hydrothermal gasification (cHTG) of biomass. S-absorption performance of MOx/C was evaluated by reacting the materials with diethyl disulfide at HTG conditions (450 °C, 30 MPa, 15 min). Their S-absorption capacity followed the order CuOx/C > CeOx/C ≈ ZnO/C > MnOx/C > FeOx/C. S was absorbed in the first four through the formation of Cu1.8S, Ce2S3, ZnS, and MnS, respectively, with a capacity of 0.17, 0.12, 0.11, and 0.09 molS molM−1. The structure of MOx/C (M = Zn, Cu, Mn) evolved significantly during S-absorption reaction, with the formation of larger agglomerates and separation of MOx particles from porous C. The formation of ZnS NPs and their aggregation in place of hexagonal ZnO crystals indicate a dissolution/precipitation mechanism. Note that aggregated ZnS NPs barely sinter under these conditions. Cu(0) showed a preferential sulfidation over Cu2O, the sulfidation of the latter seemingly following the same mechanism as for ZnO. In contrast, FeOx/C and CeOx/C showed remarkable structural stability with their NPs well-dispersed within the C matrix after reaction. MOx dissolution in water (from liquid to supercritical state) was modeled and a correlation between solubility and particle growth was found, comforting the hypothesis of the importance of an Ostwald ripening mechanism. CeOx/C with high structural stability and promising S-absorption capacity was suggested as a promising bulk absorbent for sulfides in cHTG of biomass.

www.nature.com/scientificreports/ Elliott et al. firstly introduced the application of an affordable sulfur scrubbing bed installed upstream of the Ru/C catalyst for continuous cHTG at 350 °C and 20 MPa, with different microalgae species as the feeds 13 . Raney nickel was used as an S-absorption material, which, however, presented insufficient stream desulfurization abilities. Nickel is also generally known to have low stability under reductive supercritical water conditions and suffers from sintering and leaching 9 . Peng et al. used a commercial ZnO-based S-absorbent to protect the Ru/C (5% Ru) catalyst during a continuous cHTG campaign with Chlorella Vulgaris 14 . The S-rich microalgae feed was successfully gasified to a methane-rich gas during the first 60 h time on stream, reflecting the good performance of the S-absorption by the ZnO material. But a significant accumulation of Zn-based debris was observed on the downstream Ru/C catalyst 15 , which was assumed to be a contributory cause of the catalyst deactivation after 60 h. The instability of the ZnO's binder (alumina) was proved by XRD, which underwent drastic phase transformations after SCW exposure, resulting in significant decrease of the specific surface area and pore volume of this commercial S-absorbent material. A stable binder material is desired which would preserve the integrity of the S-absorbent grain and prevent the loss of the S-absorbing phase (metal oxide) from streaming down to the catalyst bed.
In supercritical water, only a few oxides show long-term stability. Indeed, mostly refractory oxides (α-Al 2 O 3 , monoclinic ZrO 2 , rutile TiO 2 ) proved to withstand such harsh conditions, where other phases or other oxides showed phase transformation, porosity collapse, or dissolution 9,10,16 . Aside from oxides, carbon is one of the few materials that show long-term stability of its porous framework and particle integrity under SCW 9,17,18 , making it a promising low-cost binder material for such conditions. Apart from ZnO, the oxides of molybdenum (Mo), manganese (Mn), and cerium (Ce) also showed good sulfur absorption capacity with intermediate activity in decreasing organosulfur concentration, and Cu(0) demonstrated outstanding performance for both desulfurization and S-absorption 2,[19][20][21] . It is therefore expected that a composite of porous carbon and high loading of metal (or metal oxides) will be an ideal S-absorbent, showing both high S-absorption capacity and structural stability under SCW.
For the synthesis of carbon-supported metal oxide materials, a variety of methods have been developed, including physical mixing and wet chemical processes. Sharma et al. developed a simple ion-exchange method for preparing porous NiO/C catalysts for cHTG. This catalyst showed high activity and sintering resistance even after 50 h of reaction time, compared to other NiO/C catalysts reported in the literature 20 .
In this publication, we developed the synthesis of MO x /C (M = Zn, Cu, Fe, Mn, and Ce) sulfur-absorbing materials to achieve a high loading of the active phase MO x (high absorption capacity) nanoparticles (NPs) and a porous C framework. Batch tests were then performed to assess their stability in SCW and S-absorption capacity using an organic S-rich aqueous solution under cHTG conditions. The evolution of the materials upon the two treatments was carefully assessed and the thermodynamic solubility of the stable MO x phase was modeled in order to get insights in material aging, S-absorption capacity and sulfidation mechanisms.

Synthesis of MO x /C S-absorbents.
The MO x /C material preparation method was inspired by Sharma et al. 20 . The method was adjusted to be applied to various metals (M = Zn, Cu, Fe, Mn, Ce) and to obtain higher porosity and stability (pyrolysis under CO 2 or H 2 O partial pressure). Materials were synthesized generally via two steps: ion exchange followed by pyrolytic carbonization, as illustrated in Fig. 1. The loading of M (or MO x ) in the carbon matrix relies on the thermodynamics of ion exchange between protons from the carboxylic acid cation exchange resin and M precursors in the first step. Since carboxylic acids are weak acids with small ionization constants, neutral to alkaline media facilitates the cation exchange process. For this purpose, ammonium hydroxide was used to regulate the pH, which specifically made Zn and Cu more soluble in alkaline solution by forming [Zn(NH 3 ) 4 ] 2+ and [Cu(NH 3 ) 4 ] 2+ complex ions. However, Fe, Mn, and Ce could not form soluble ammonia complexes, their solubility being low in alkaline conditions (hydroxide precipitation). To achieve high M loading, the synthesis methods of different MO x /C materials varied according to the solubility of M.
ZnO/C and CuO x /C. A 0.2 mol L −1 Zn 2+ or Cu 2+ precursor solution was prepared by dissolving zinc acetate dihydrate (98.0-101%, Alfa Aesar) or copper acetate monohydrate (98%, Alfa Aesar) into de-ionized (DI) water, followed by adding ammonium hydroxide solution (28-30%, Sigma) to reach a pH of 10. The ion exchange resin (DIAION WK 11, Mitsubishi Chemical) used has the carboxylic acid group (-COOH) as the exchange group, with the capacity of 3.0 meq mL −1 resin and a measured volumetric weight of 0.453 g mL −1 , which corresponds to 6.623 mmol g −1 . The resin was then weighed and added to the alkaline [Zn(NH 3   FeO x /C, MnO x /C, and CeO x /C. Fe 2+ , Mn 2+ and Ce 3+ ions have low solubility in alkaline solution due to their precipitation as hydroxides. Moreover, they could not be converted to soluble complexes by coordination with ammonia. Thus, the pH of their precursor solutions could not be increased to a value above 8, but was limited by the starting point of the precipitation when increasing the pH. Iron(II) acetate (95%, Sigma-Aldrich), Manganese(II) acetate (tetrahydrate, 99%, Sigma-Aldrich), and Cerium nitrate (hexahydrate, > 99%, Fluka) were used as the precursors of Fe, Mn, and Ce respectively. Fe(II) and Mn(II) precursors were chosen for their solubility in higher pH solutions as opposed to their counterparts with higher oxidation states. The initial pH of the 0.2 mol L −1 solutions were 6, 7.5, and 4 for Fe 2+ , Mn 2+ , and Ce 3+ , respectively. NH 3 was added to the three metal solutions dropwise to adjust the pH to 8, 7.5, and 6, respectively, which was the maximum value allowed without precipitation. The following ion exchange and pyrolysis processes were the same for Fe, Mn, and Ce as with Zn. An inert atmosphere (N 2 ) was used for Fe and Mn during the ion exchange to avoid the oxidation of their divalent cations.

Reference porous C.
To investigate the effect of the carbon support, a reference porous C was also tested as a control. The preparation of this C material was carried out by acid washing of the as-prepared ZnO/C illustrated in "ZnO/C and CuO x /C" section, allowing full dissolution of the ZnO. In detail, for preparing 0.5 g of carbon material, 1.72 g of ZnO/C (71 wt.% ZnO) was dispersed in 0.5 L 0.1 M HCl solution at room temperature with mild stirring (200 rpm) for 24 h. After separation from the acid by filtration, the spherical particles were washed with plenty of DI water (5 × 1 L) to remove the remaining Cl − and then dried at 110 °C to obtain the as-prepared porous C.

Characterization of MO x /C S-absorbents.
The specific surface area (SSA) and pore structure of fresh and spent MO x /C S-absorbents were determined by measuring the N 2 -physisorption (77 K) isotherm using a Quantachrome Autosorb AS1 instrument. The samples were outgassed in a dynamic vacuum for a minimum of 5 h at 250 °C. Brunauer-Emmett-Teller (BET) theory was applied to determine the SSA. The total pore volume was determined at relative pressures p p 0 -1 ≥ 0.99. The micropore volume (V MP ) was determined by the t-method developed by Lippens and de Boer 22 . The non-micropore volume is defined here as the external volume (V Ext ) and is calculated by subtracting the micropore volume from the total pore volume: V Ext = V Tot − V MP . The external surface area (SSA Ext ) that defines the sample surface area without the micropore contribution was calculated by the same method. The pore diameter distribution and the mean pore diameter were calculated using the adsorption branch based on the Barrett-Joyner-Halenda (BJH) method.
The crystalline phase structure of the solid materials was determined by powder X-ray diffraction (XRD) on a Bruker D8 Advance diffractometer equipped with Ni-filtered Cu Kα-radiation (λ = 1.5418 Å). The diffraction spectrum was collected in the 2θ range from 5° to 90° with the acquisition interval of 0.03° per every 4 s.
To study the material morphology, particle size and shape, transmission electron microscopy (TEM) was performed on two microscopes: a JEOL JEM 2010 microscope operated at 200 keV with a LaB 6 electron source equipped with a slow-scan CCD camera (4008 × 2672 pixels, Orius Gatan Inc.) and a probe corrected JEOL JEM-ARM200F (NeoARM) microscope equipped with a cold field emission gun operated at 200 keV and a OneView CMOS camera (IS-version; Gatan). The latter microscope was also used for STEM and EDX investigations, for which a JEOL EDX detector "Centurio 100" was used to measure the elemental composition. The particle size of MO x /C in TEM images was measured using ImageJ software, 100 particles were randomly selected from representative images.
Quantitative elemental analysis of solid and liquid samples was performed on an ICP-OES device (Agilent 715). The analyzer was calibrated with different dilutions of a multi-element standard solution prepared from certified standards. The standards of the individual elements were purchased either from Ultra Scientific or from Merck. Liquid and solid samples were all pre-treated by digestion in acid containing 5 mL 65% HNO 3 and 1 mL 37% HCl.
Evaluation of MO x /C S-absorbents under HTG conditions. The performance of MO x /C materials was evaluated by using an unstirred 316 stainless-steel batch reactor (BR) with 28 mL volume, assembled using components from HiP. More details of the reactor setup were given in previous work 23 . A preheated fluidized sand bath (IFB51, Techne) was used to quickly heat the BR to the targeted temperature of 450 °C within ca. 15 min. The total volume of the water-based feed was always set at 6.7 mL, in order to reach 30 MPa at 450 °C. The reaction time for all the experiments was kept at 15 min from when the BR temperature reached 450 °C. After each BR test, the reactor was quenched by immersion into a cold-water bath until reaching room temperature. Then, the BR was depressurized and the discharged gas was sampled in a 3 L gas bag (Multi-Layer Foil, RESTEK) for immediate offline gas analysis using a calibrated µGC ( www.nature.com/scientificreports/ 4 mL of isopropanol was added to the opened BR to dissolve the water-insoluble organics (solids, liquid organic phase), allowing the sampling of a homogeneous liquid. The sampling of spent solids was carried out by direct separation from the liquid, followed by washing the particles with ethanol and water (three times each) before drying at 110 °C.
To study the individual effect of pure SCW on MO x /C (M = Zn and Cu) materials, an "SCW-stability" test was carried out by exposure of 100 mg MO x /C (M = Zn and Cu) to 6.7 mL DI water in the BR and maintaining the HTG conditions of 450 °C and 30 MPa for 24 h.
The sulfur absorption performance of the MO x /C materials under HTG conditions was evaluated by using aqueous hydrocarbon as the feed to mimic the wet biomass and using model organoS (organic sulfur compound) as the S source. To evaluate the capacity of MO x /C materials to absorb sulfur species, an excess of S with regard to M was used. M/S ratio was set to 2 (mol M mol S −1 ) considering some metals (e.g. Cu) can form an M 2 S composite with S. S should be in the form of H 2 S or a form leading to the formation of H 2 S in SCW. To that end, diethyl disulfide (DEDS) was used to generate H 2 S in SCW as it is known to have low stability under these conditions and to lead to H 2 S formation in situ 24 . With the BR setup and reaction conditions of 450 °C, 30 MPa, and 15 min, 6.7 mL of model S feed solution (DEDS in a water/isopropanol mix) was used corresponding to an absolute sulfur amount of 0.397 mmol S . The MO x /C was added to the DEDS solution in a quantity such that the M amount was 0.794 mmol M . The amount of C tested was set at 100 mg to keep a similar total weight as with the porous MO x /C (M = Zn, Fe and Cu). To validate the reproducibility of such S-absorption tests, the tests without solid material ("solution alone"), with ZnO/C, and with CuO x /C were repeated with a coefficient of variation of less than 6.5%. The corresponding standard deviation are indicated in Fig. 3. Due to the optimum performance of CuO x /C, a "maximum" S-absorption capacity test was individually carried out for it using the same DEDS feed but with different conditions of M/S = 1 mol M mol S −1 with an HTG reaction time of 15 h.
The analysis of sulfur compounds. The total amount of S in either liquid phase or solid phase was determined by ICP-OES with the instrument and sample pretreatment methods described above in "Characterization of MO x /C S-absorbents" section. An ion chromatography device (IC, Metrohm) using a Metrosep A Supp 10-100/4.0 column and Metrohm 732 IC conductivity detector was used to quantify the sulfate ions in liquid samples. Light organosulfur compounds in the liquid residue were analyzed by gas chromatography (GC, Agilent 7890A) using an Agilent J&W DB-Sulfur SCD column (40 m × 0.32 mm ID) with helium (quality 6.0) as the carrier gas. A sulfur chemiluminescence detector (SCD, Agilent 355) with a dual plasma burner was coupled for the selective analysis of S-containing organic compounds 25 . An identical column equipped in parallel and connected to a flame ionization detector (FID) was used to give general information on the presence of other (S-free) organic compounds. The oven temperature was programmed as 40 °C for 7 min before rising at 7 °C min −1 up to 220 °C for 8 min. Theoretically, the quantitative calibration of sulfur compounds can be based on the "equimolar response" of SCD 26 . Dibenzothiophene was used as an internal standard to perform a quantitative analysis of volatile sulfur compounds. The sum of all the organosulfur calculated by this semi-quantitative method is always higher than the total amount of S in liquid phase measured by ICP-OES. Considering a higher accuracy of the latter, the amount of organosulfur compounds was proportionally corrected based on the total S measured by ICP-OES. The retention time (RT) of S compounds was determined by testing various pure organosulfur standards, including methanethiol, dimethyl sulfide, dimethyl disulfide (DMDS), dimethyl trisulfide (DMTS). Thanks to the column and method used, a linear relationship existed between the boiling point (BP) and RT and is given in Fig. S1, so that the BP of an unknown sulfur-containing compound could be extracted from the RT.  Table S2, to form the corresponding metal ions or hydrated metal ions, were used to model the dissolution of MO x in pure neutral water. Redox reactions of ZnO, Cu 2 O, Fe 3 O 4 , MnO, and CeO 2 were not considered because of their stable oxidation states under HTG conditions, proven by the material characterizations in the experimental work of this study. Only the dissolution of Cu(0) was considered to be a redox reaction with the formation of Cu + and aqueous H 2 . To simplify the model, Cu + was regarded as the preferential dissolution product of Cu(0), Cu 2+ being excluded due to the lower thermodynamic barrier for its formation. Eight different conditions (T, P, ρ H2O ) were modeled, as listed in Table S4, corresponding to different states of water from liquid to supercritical, which were extracted from a representative T&P profile (see Fig. S12) in the experimental work of this paper. Following the revised Helgeson-Kirkham-Flowers (R-HKF) thermodynamic model applied by Shock et al. 27 and Jocz et al. 16 , equilibrium constants (K eq ) of the MO x dissolution reactions listed in Table S2 were firstly calculated. Secondly, concentrations of dissolved M ions or hydrated M ions in water as multiple unknowns were solved using Matlab. All the calculation details are given in the Appendix of the Supplementary Information.
Briefly, K eq was calculated from the change in Gibbs free energy of the dissolution reactions (ΔG rxn ) as shown in Equation S1, where the calculation of ΔG rxn was based on the apparent standard partial molar Gibbs free energy of the formation �G f T, ρ H 2 O j of each species (j) involved in the reaction and their stoichiometric coefficient ( v j ), given in Equation S2. Depending on the phase of j, �G f T, ρ H 2 O j was calculated using the specific expression.
�G f T, ρ H 2 O j of solid j and H 2 O under a targeted condition (T, P, ρ H2O ) was calculated using the differential expression for apparent standard partial molar Gibbs free energy (Equation S3). The thermodynamic parameters at the standard reference temperature (T 0 = 298.15 K) and pressure (P 0 = 1 bar) for such calculation, such as the standard partial molar Gibbs free energy of formation of j (ΔG j 0 ) from the elements in their stable form, and the www.nature.com/scientificreports/ standard molar entropy (S j 0 ) were obtained from the thermochemical database 28,29 . Table S3 summarizes those standard thermodynamic parameters of MO x solid species. Table S4 lists the calculated thermodynamic and solvent-related parameters of water at the selected eight different conditions. �G f T, ρ H 2 O j of ions and dissolved H 2 and O 2 was calculated using the R-HKF equation 27 (Equation S5). The standard thermodynamic and HKF parameters (species-dependent non-solvation parameters) of specific aqueous ions were obtained from Shock et al. 30 and are reprinted in Table S5. �G f T, ρ H 2 O j of H + was set as the reference for all aqueous species and equaled to 0 at all conditions, while �G f T, ρ H 2 O j of OH − was calculated using Equation S14.
Therefore, based on the Equations S2-13, ΔG rxn and K eq of the targeted dissolution reactions of MO x under the eight different states of water were calculated as listed in Table S6. To further calculate the molar concentrations m j (mol kg −1 ) of all the dissolved species in water at a specific condition, Equation S15 relating the equilibrium constants (K eq ) to the thermodynamic activity coefficient (γ j ) was used, where γ j could be correlated to the ionic strength (I) and the Debye-Hückel parameter (A Φ ) using Davies extension of the Debye-Hückel equation 31 in Equation S16. Assuming N total aqueous species (H + , OH − , and N-2 dissolved ions from the MO x ) are existing in water, there are N + 1 unknown variables (N concentrations m j plus ionic strength I). Combining the N-2 equilibrium expressions (Equation S15) for all dissolution reactions, the definition of the ionic strength (I) of the solution (Equation S18), the equilibrium equation for the ion product of water, and the expression of charge neutrality (Equation S19), N + 1 equations were set up to solve the N + 1 unknowns. Matlab's nonlinear leastsquares solver function (lsqnonlin) with the "Levenberg-Marquardt" algorithm was used to solve the equations and ultimately calculate a total concentration of dissolved compounds.

Results and discussions
Synthesis development and characterization of MO x /C materials. Aiming at high MO x loadings in the MO x /C composites as well as a uniform distribution of MO x in the C matrix, the synthesis process was optimized in terms of ion exchange and pyrolytic carbonization as illustrated in Fig. 1.
The optimization of pyrolysis conditions in ZnO/C synthesis. In a previous study, pure Ar was used for the pyrolytic carbonization of the methacrylic acid resin 32 . For optimum sulfur absorption performance, one should aim at increasing the porosity of the MO x /C material and increasing MO x loading (more C oxidized). For that, weak oxidants, i.e. H 2 O and CO 2 , were added to Ar in the pyrolysis process. Steam or CO 2 co-feeding with the inert gas stream was proved in biomass pyrolysis to enhance the porosity of the biochar 33 . This treatment with steam or CO 2 supposedly also favors the formation of phases with higher stability in supercritical water. Figure S2 presents the XRD patterns of two ZnO/C materials synthesized with 1.4 vol.% H 2 O in Ar and 1 vol.% CO 2 in Ar. They present very similar crystalline composition, all diffraction peaks could be assigned to hexagonal ZnO. Similar peak intensity was also observed, and the average ZnO crystallite size according to the Scherrer Equation was found to be 22 nm and 21 nm for "CO 2 in Ar" and "H 2 O in Ar", respectively.
Different pyrolysis temperatures (400, 500, and 600 °C) were also studied with the H 2 O-and CO 2 -containing Ar. A control test with pure argon at 500 °C was also performed to investigate the effect of water or carbon dioxide. The ZnO loading and the pore properties of the ZnO/C materials obtained under different pyrolysis conditions are summarized in Table 1.
As shown in Table 1, under the same pyrolysis temperature of 500 °C, the presence of H 2 O or CO 2 in the Ar increases the ZnO content in the as-prepared ZnO/C. In parallel, the specific surface area and pore volume seem to be lower. With H 2 O/Ar or with CO 2 /Ar, the ZnO content increased with the pyrolysis temperature, indicating the gasification of carbon. Eventually, at the highest temperature tested, i.e. 600 °C, the material shows a heterogeneous appearance with the presence of white particles among the black/grey particles, indicating a nearly complete gasification of the carbon in some area of the material bed. Even though 600 °C produced the highest surface area, pore volume and loading both with H 2 O/Ar and CO 2 /Ar, the heterogeneity of the MO x distribution in the material led to choose 500 °C under H 2 O/Ar atmosphere as optimum pyrolysis conditions. The use of H 2 O leads to a loading 9 wt% higher than when CO 2 was used, while the pore volume are > 0.15 cm g −1 in both www.nature.com/scientificreports/ cases. The optimal pyrolysis condition was therefore defined as 500 °C and 1.4 vol.% H 2 O in Ar. These conditions were used for the synthesis process of all other MO x /C materials.
Characterization of the as-prepared MO x /C materials. As shown in Table 2, both ZnO/C and CuO x /C reached a high metal loading with > 0.3 mol M mol C −1 (71 and 62 wt%, respectively), due to the high solubility of metal cations in alkaline solution. The maximum pH allowed before precipitation for Fe 2+ , Mn 2+ , and Ce 3+ solutions was 8, 7.5, and 6, respectively. Higher pH is favored during the ion exchange step in the MO x /C synthesis. Thus their MO x loadings follow the order of FeO x /C (56.4 wt%) > MnO x /C (50.0 wt%) > CeO x /C (27.3 wt%). Their crystalline phases were analyzed by XRD (Fig. S4) and are summarized in Table 2. ZnO/C, FeO x /C and CeO x /C present single-phase MO x crystalline grain, which is ZnO (hexagonal), Fe 3 O 4 (cubic), and CeO 2 (cubic), respectively. Differently, CuO x /C and MnO x /C both show more than two crystalline phases of metal oxides. For CuO x , a mixture of Cu 2 O (cubic) and Cu (cubic) is observed according to the XRD pattern (Fig. S4b), which is different to the pure Cu(0) phase obtained by using dry Ar to pyrolyze [Cu(NH 3 ) 4 ] 2+ exchanged methacrylic acid-based resin 32 . MnO x (Fig. S4d) shows the main crystalline phase as MnO (cubic) with small amounts of Mn 2 O 3 (orthorhombic).
The as-prepared materials all showed nano-scale particles uniformly dispersed in the carbon matrix as evidenced by TEM in Fig. 2. The particle size distribution (PSD) varied by the type of M, which were evaluated from TEM images and are plotted in Fig. S5. From the representative images given in Fig. 2a-e, the particle size of ZnO, CuO x , FeO x , MnO x , and CeO x embedded in C is in the range of 14 ± 5, 26 ± 7, 15 ± 4, 20 ± 17, and < 5 nm, respectively, following the order CuO x > MnO x > FeO x ≈ ZnO > > CeO x . Their crystalline phases and corresponding crystallite size determined by XRD (in Fig. S4) are ZnO (19 nm), Cu 2 O (47 nm)/Cu (53 nm), Fe 3 O 4 (20 nm), MnO (45 nm)/Mn 2 O 3 (46 nm), and CeO 2 (4 nm) indicating overall that crystalline particles are larger than the average. The different particle and crystallite size of specific MO x formed might be explained by Tammann temperatures (summarized in Table 2) that correlate to a material resistance towards sintering 34 . ZnO, FeO x , and CeO x present finer and more uniform particle sizes than CuO x and MnO x thanks to the higher Tammann temperatures and purer oxidation states. TEM micrograph of the control C material prepared by acid washing of ZnO/C is shown in Fig. 2f. The absence of particles supports the ICP analysis in that no Zn remained after the acid washing and indicates a high purity of C. N 2 adsorption-desorption isotherms (Fig. S3) were used to characterize the pore structure of MO x /C and C with the results summarized in Table 2. ZnO/C, CuO x /C, and FeO x /C all show porous properties with SSA higher than 200 m 2 g −1 and total pore volume over 0.14 cm 3 g −1 . Their isotherms indicate mainly type I/II with little to no hysteresis indicating an open pore structure with some microporosity 36 . They have a broad pore size distribution, as presented in the inserts of Fig. S3a-c. ZnO/C, CuO x /C, and FeO x /C have a microporosity of 47%, 50%, and 11% of the total pore volume, respectively. The mean pore size of the three materials is in the range of 2-4 nm, which is much smaller than the particle size of MO x embedded into the carbon matrix.
The C material prepared by acid washing of ZnO/C shows a specific surface area of 1077 m 2 g −1 and a total pore volume of 1.0 cm 3 g −1 . Its isotherms (see Fig. S3f) present a continuously increasing adsorption branch in line with a broad pore size distribution, but a very sharp drop of adsorbed gas at P/P 0 = ca. 0.47 on the desorption branch, which is typical of mesopores connected to the outer surface through micropores (ink-bottleshaped pores). The pore size distribution obviously indicates two pore families: micropores (volume 0.25 cm 3 g −1 ) and mesopores mainly distributed around 8 nm. The TEM image (see Fig. 2f) points to a very homogeneous porous material with a high porosity in line with a dissolution of ZnO particles yielding mesoporous cavities. www.nature.com/scientificreports/ The ink-bottle-shaped pores confirm that the ZnO/C material consists mainly of ZnO particles embedded by microporous carbon, which after dissolution in acid leaves cavities connected to the outer surface through micropores. When metal-free resin was pyrolyzed, the product obtain consisted of non-porous carbon block/sheets, indicating that the polymer melted before pyrolysis. In all cases, the metal content impacted the resin melting/ pyrolysis as all MO x /C consisted of spheres of the same diameter as the starting resin. However, MnO x /C and CeO x /C show poor porosity with a specific surface area below 3 m 2 g −1 . This indicates that the two elements do not impact the decomposition of the polymer the way Zn, Cu, Fe or Ni do, maybe because the salts decompose and the oxides form after the polymer melted and started to pyrolyze 20 . In the case of Ce, the low pore volume and surface area might also be explained by the lower loading of Ce in the material.  www.nature.com/scientificreports/ As illustrated in Fig. 3a, out of the 397 µmol S of DEDS in the feed, after cHTG reaction without S-absorbents (solution alone), about 22% of the sulfur remained in solution as volatile organoS. A small amount (ca. 15 µmol S ) of SO 4 2− was also detected in all the spent liquids but most likely originated from the oxidation of dissolved H 2 S by air in the aqueous solution upon sample preparation 37 . The feed only contains two organic compounds, i.e. isopropanol ((CH 3 ) 2 CHOH) and DEDS (C 2 H 5 SSC 2 H 5 ), SCW should break down most C-C and C-S bonds such that ethanethiol (C 2 H 5 SH) should be the main remaining organoS in all spent liquids 38 .
Identified and semi-quantified by GC-SCD (see Fig. S6b), the liquid phase after testing was mainly made up of volatile sulfur compounds (BP ≤ 350 °C). Without using S-absorbent, the spent liquid was mainly composed of 60 µmol S ethanethiol and 15 µmol S DEDS. Note that the DEDS observed might be formed back from ethyl thiyl radicals during the quenching of the reactor, or from ethanethiol after sampling 19 . Other compounds such as isopropyl disulfide ((CH 3 ) 2 CHSSCH(CH 3 ) 2 ), methyl ethyl trisulfide (CH 3 SSSC 2 H 5 ), and diethyl trisulfide (C 2 H 5 SSSC 2 H 5 ) might also have been formed from corresponding radicals present in the supercritical fluid or from the corresponding thiols and H 2 S after sampling. The presence of different materials in the solution only led to a significant decrease of ethanethiol, their impact on the concentration of other identified volatile organosulfur compounds was marginal except for ZnO/C, which showed a higher concentration of trisulfides.
Comparing the amount of S absorbed in the materials (striped pattern in Fig. 3a), it is interesting to note firstly that the C alone, when exposed to HTG conditions, leads to an accumulation of S (see Fig. 3a,b, "C"). , where n(S absorbed in MO x /C) are the moles of S absorbed in MO x /C obtained experimentally, n S absorbed in micropore is the product of MO x /C micropore volume (given in Table 2) and the micropore S-absorption capacity of C (1.81 mmol S cm micropore −3 calculated by the S-absorption capacity test of C). n(M) are the moles of metal used for the test. c Not detected. d Not applicable. www.nature.com/scientificreports/ Micropores are known to favor coking and quickly foul during cHTG reaction which might explain the deposition of sulfur on the material in the form of coke, or simply adsorbed inside 39 . Although the porous framework was maintained in the spent C with a total SSA of 599 m 2 g −1 (Fig. S7a), the micropore volume decreased by 64% compared to the fresh C. A comparison of the pore size distribution between the fresh C and spent C directly shows the decreased micropore proportion after the reaction (Fig. S7b) supporting this hypothesis. Accordingly, the S-absorption capacity in mol S mol M −1 (Table 3 and Fig. 3b, left Y-axis) is a corrected value after subtracting the micropore-absorbed S using the micropore volume of MO x /C (Table 2) and the microporous capacity of C (1.81 mmol S cm micropore −3 calculated from the test of C). The order of S-absorption capacity (mol S mol M −1 ) can be summarized as CuO x /C > CeO x /C ≈ ZnO/C > MnO x /C > FeO x /C. CuO x /C presented the highest S-absorption capacity with 0.17 mol S mol M −1 and 0.048 g S g absorbent −1 . CeO x /C shows a slightly higher absorption capacity than ZnO/C, 0.12 vs. 0.11 mol S mol M −1 (cf. Table 3), even though the absolute amount of S captured by ZnO/C is higher than CeO x /C (Fig. 3a). In the case of ZnO/C, up to 12% of the deposited S might be due to the carbon microporosity alone. The very low surface area and microporosity of CeO x /C should have resulted in a negligible amount of S deposition on the carbon itself, which explains, together with the low loading of ceria, the low massbased S-absorption capacity (only 0.006 g S g absorbent −1 ) in Fig. 3b. FeO x /C shows 48 µmol sulfur captured, very similar to the C alone, but because FeO x /C had five and ten times lower total surface area and micropore volume, respectively, it can be estimated that the material absorbed 0.06 mol S mol M −1 (see Table 3) after correction for the estimated C absorption. The capacity of MnO x /C to absorb S seems also high, taking into consideration its low surface area and the probably negligible impact of its carbon binder on the material performance. A "maximum" S-absorption capacity test was carried out solely on CuO x /C under a much longer reaction time (15 h) but with a lower M/S ratio (1 mol M mol S −1 ). After this long test, a higher S-absorption capacity was reached, that is 0.43 mol S mol Cu −1 , compared to 0.17 mol S mol Cu −1 after 15 min with M/S ratio = 2 mol M mol S −1 . The inability of copper to reach full sulfidation might be due to the formation of H 2 S in the gas phase before it can react with the S-absorbent, which, due to the absence of stirring in the reactor, will only slowly diffuse to the supercritical fluid and to the solid material.

ZnO/C CuO x /C FeO x /C MnO x /C CeO x /C CuO x /C*
The XRD patterns of MO x /C after these sulfur absorption capacity tests are shown in Figs. 4, 5, 6, 7, 8, and are summarized in Fig. S4. The original oxidation state of metals in those MO x /C materials was mostly maintained. More specifically, ZnO, Fe 3 O 4 , Cu 2 O, MnO and CeO 2 were the dominating crystalline phases in the fresh ZnO/C, FeO x /C, CuO x /C, MnO x /C, and CeO x /C, respectively, and did not suffer from reduction/oxidation during reductive HTG reaction. The spent MO x /C all show crystalline metal-sulfide phases with the exception of FeO x /C, which is in line with the very limited S-absorption performance of iron described in an earlier work 19 . In spent MO x /C (M = Zn, Cu, Mn, and Ce), crystalline sulfides are formed as ZnS, Cu 1.8 S, MnS, and Ce 2 S 3 , respectively, revealing the theoretical S-absorption capacity of those materials. Combining with the experimental S-absorption capacity presented in Fig. 3b, the sulfidation rate of those S-absorbents were evaluated and are listed in Table 3. In terms of S-absorption efficiency, CuO x /C shows much better performance than ZnO/C, MnO x /C, and CeO x /C. The material stability of those S-absorbents under such reaction conditions is another important aspect, which was evaluated by material characterization in the next section.

Different structural evolution of MO x /C materials under HTG conditions. Evolution of ZnO/C
by SCW and upon sulfidation under HTG condition. XRD and representative TEM and EDX results for asprepared, SCW-treated, and sulfided ZnO/C are presented in Fig. 4.
The effect of SCW. As mentioned in "Characterization of the as-prepared MO x /C materials" section, the asprepared ZnO/C (Fig. 4a) is composed of ZnO NPs with the particle size 14 ± 5 nm, homogeneously distributed in a carbon matrix (Fig. S5a). After exposure to SCW (water at 450 °C and 30 MPa) for 24 h, ZnO particles migrated outward from the C matrix along with a size increase of the particles as shown in Fig. 4b, in line with the increased crystallite size from 19 to 29 nm indicated by their respective XRD patterns (Fig. 4d). Little literature exists on the mechanisms of particle growth from a supported porous material under supercritical water conditions. The high surface energy and thermodynamic potential inherent to nanoparticles are a driving force for nanoparticle growth, and they are considerably worsened by supercritical water conditions. Coarsening typically occurs via Ostwald ripening, by migration of the particle and their coalescence, or by a combination of both phenomena 40 .
In the case of Ostwald ripening, particles would be immobile in the carbon matrix and monomers of zinc oxide would migrate from particles below a critical size to larger ones. This phenomenon can occur by the dissolution of monomers and/or via surface migration. The variation of ZnO solubility in water from standard conditions to SCW state (450 °C, 30 MPa, ρ H2O = 0.144 g mL −1 ) can be one reason for such ripening and has been explored in various studies 41,42 . Thermodynamic modeling in the next "Thermodynamic modeling of MO x dissolution in water" section will estimate the solubility of ZnO in water along the heating-up process from liquid water (50 °C, 3.5 MPa) to supercritical water (450 °C, 30 MPa). This variation of solubility upon heating (and cooling) under pressurized water might favor particle growth by Ostwald ripening.
Another coarsening mechanism is through the migration of particles and their subsequent coalescence. The majority of the work done on particle aggregation in supercritical water covers salt precipitation or particle synthesis, which are dominated by supersaturation of an ionic solution 43,44 . Small clusters with large charge imbalance were calculated by molecular dynamics to be readily formed and to dominate the growth of FeCl 2 clusters 45 . Low Tammann point, i.e. close to the operation temperature, favors particle migration and coalescence, but in the case of ZnO the Tammann point is rather high (850 °C).
More specifically, several parameters influence particle migration, such as ionic strength, cation species or pH, but also the structural hydration layer thickness of nanoparticles that increases with decreasing particle  46 . Under SCW conditions, it was shown that the thickness of the hydration layer increases with temperature, doubling from 400 to 500 °C, while an increase of pressure led to a decrease of that layer 46 . It can be considered that a hydration layer might be favored on zinc oxide under the reaction conditions used in this work. In pure supercritical water, both Ostwald ripening and particle migration/coalescence might be involved in the growth of particle size. Many coarsened ZnO particles present a nearly defect-free rod-like morphology, indicating the growth of hexagonal ZnO crystallite in the direction [001].
Evolution upon sulfidation under HTG condition. The ZnO agglomerates and rods are also found in the spent ZnO/C after the S-absorption capacity test with DEDS (Fig. 4c), which caused the broadening of PSD throughout the material (Fig. S5a). ZnO was not fully sulfided because of a stoichiometric excess of ZnO in the feed (M/S = 2 mol M mol S −1 ). The lattice space ZnO [100] of a sidewall of one rod was measured from an HR-TEM image (Fig. S8), which confirms the length-wise growth of hexagonal ZnO along its perpendicular direction [001] 47 . STEM-EDX elemental mapping indicates an irregular shape of ZnS particles with a size up to 400 nm and a representative Zn/S atom percentage to be roughly 50/40. More importantly, the distribution of S throughout the sample was very heterogeneous, with some Zn particles having an O:S ratio of 1, while other particles had an O:S close to 0 (see Fig. 4c). Overall, EDX data indicate that S combines with Zn by replacing O, confirming zinc oxide sulfidation, as expected from exposing this material to H 2 S (from DEDS decomposition 24 ).
Trisulfide radical ion S 3 ⋅− was evidenced to be a ubiquitous sulfur species in reductive S-rich hydrothermal fluids, and has been shown to coordinate with Pt(II), Pt(IV), and Au(I) to form soluble and stable Pt II (HS) 2 (S 3 ) 2 2− , Pt IV (HS) 3 (H 2 O)(S 3 ) 2 − , and Au(HS)S 3 − in hydrothermal fluids 48,49 . These complexes probably explain the migration of elements in the Earth's crust and the formation of ores rich in these metals. It is hypothesized that all chalcophile elements such as Zn(II) might form similar hydrated trisulfide complexes in reductive SCW 50 . Such complexes would hence favor particle growth of ZnS particles, e.g. by Ostwald ripening, but the absence of large ZnS crystals (Fig. 4d) and the presence of small ZnS particles aggregated in defined structure, e.g. rod-like (Fig. 4c), indicate that this mechanism is not favored. Similar to the conclusion from Tiemann et al. 51 who modeled the kinetics of ZnS NPs growth at standard T&P, the mechanism of ZnS growth corresponds to coalescence with barrier-controlled attachment rather than Ostwald ripening. Indeed, Ostwald ripening involves material dissolution that is less favored for ZnS (K sp = 10 -24 ) than for ZnO (K sp = 10 -16 for Zn(OH) 2 ) 52 .
Detailed STEM-EDX study of a sulfided rod-like particle (Fig. 4c, top right) indicates a particle with a rough surface composed of aggregated 3-5 nm ZnS NPs and two ~ 50 nm partially crystalline ZnO particles at the tip of the needle. First, this observation indicates that crystallization of ZnS particles are not favored under such conditions but that aggregation is strongly favored as no free 3-5 nm particles were observed. This is in line with the www.nature.com/scientificreports/ findings of Ma et al. who observed that sulfidation of the ZnO NPs decreases their surface charge and their Zeta potential, promoting their aggregation, compared to pristine ZnO NPs at standard pressure and temperature 52 . The structure of ZnO core covered by ZnS NPs shell resembles what Ma et al. 52 found when sulfiding ZnO NPs in NaSH aqueous solution at room temperature and pressure. These observations would hence support the hypothesis that ZnS is formed by the dissolution of Zn oxide and its precipitation in the form of ZnS. Based on the observations made and the existing literature, it can be hypothesized that during the S-absorption tests performed, ZnO coarsened probably via competitive Ostwald ripening (ZnO dissolution before ZnO monomers/nuclei migration) and particle migration/coalescence, yielding to crystalline needle-like particles and round poorly crystalline ones. In parallel to this, sulfidation of ZnO occurred likely by ZnO dissolution and then ZnS precipitation with hydrated zinc trisulfide complexes as the intermediates, yielding ZnS NPs agglomerated in place of bulk ZnO particles.
Evolution of CuO x /C by SCW and upon sulfidation under HTG condition. Figure 5 summarizes the XRD and representative TEM and EDX results for as-prepared, SCW-treated, and sulfided CuO x /C after the S-absorption capacity tests.
The effect of SCW. The as-prepared CuO x /C, as mentioned in "Characterization of the as-prepared MO x /C materials" section, consists of 26 ± 7 nm Cu 2 O/Cu NPs randomly distributed in the porous C matrix (Figs. 5a,  S5b), the SCW-treated CuO x /C (see Fig. 5b) shows a distinct separation of porous C matrix and CuO x particles with an increase of particle size. Two different morphologies of CuO x particles are presented, which are hollow nanospheres (d = 40-70 nm) and micrometric agglomerates. The corresponding XRD pattern (red curve in Fig. 5d) indicates a decrease of crystalline Cu 2 O in favor of Cu(0) when comparing SCW-treated CuO x /C to the www.nature.com/scientificreports/ as-prepared material. SCW has no redox effect on Cu 2 O and Cu(0) as evidenced by Pocock et al. 53 . This change in diffraction pattern could be explained by the reduction of copper by organic compounds formed by exposing the carbon matrix to SCW. The difference in Tammann points of Cu 2 O and Cu(0) (480 °C and 406 °C, respectively) would lead Cu(0) to sinter more readily than Cu 2 O and would also favor the formation of larger crystals after SCW at 450 °C. The decreased crystallinity of Cu 2 O could be explained by the Ostwald ripening process with crystal dissolution as the onset, supported by its higher solubility in SCW than Cu(0) 53 which will be also proved by thermodynamic modeling in the next "Thermodynamic modeling of MO x dissolution in water" section. In line with the findings by Zhang et al., the hollowing process of Cu 2 O spherical particles was driven by Ostwald ripening mostly through aqueous copper ions migration and formed ultimately poorly crystalline Cu 2 O hollow spheres 54,55 . Based on the literature and the XRD result (red curve in Fig. 5d) , the crystalline constituents of spent CuO x /C are Cu 2 O (average D p = 19 nm) and Cu 1.8 S (average D p = 31 nm) as indicated by XRD (green curve in Fig. 5d). The maintained peak intensity of Cu 2 O and the absence of Cu(0), as opposed to the SCW-treated CuO x /C, suggests a preferential sulfidation of Cu(0) rather than Cu 2 O. This is in line with the work of Ziegler et al. 56 who suggested that in the presence of hexanethiol in SCW environment, CuO firstly reduced to Cu(0) on which thiol adsorbs. Ethanethiol being an intermediate decomposition product from DEDS in SCW (cf. Fig. S6b), one would expect it to be adsorbed to Cu(0) surface before further decomposing to Cu x S and ethylene. In this work, Cu 2 O is more likely reduced to Cu(0) by isopropanol or its decomposition products. STEM-EDX study of the spent material indicates a very heterogeneous sulfidation of CuO x /C as observed with ZnO/C (see Fig. 5c). EDX results indicate that Cu-and S-rich particles are generally shapeless and often reach a size of a few hundred nanometers. The semi-quantitative EDX analysis of S-rich Cu particles suggests an atomic proportion of Cu, S, and O to be 65%, 18%, and 1%, respectively, indicating roughly a S/Cu ratio of 0.28 mol S mol Cu −1 in the selected region. Considering the theoretical S-absorption capacity of 0.56 mol S mol Cu −1 for Cu 1.8 S observed by XRD, this area shows a 50% sulfidation rate, higher than the 31% of the bulk material (Table 3) due to the presence of non-sulfided copper. The EDX analysis of a Cu-rich but S-free area indicated 32 mol%, 1 mol%, and 15 mol% of Cu, S, and O, respectively, in line with Cu 2 O. Hollow Cu 2 O particles with a diameter ranging from 30 to 70 nm are also found, like those observed on the SCW-treated CuO x /C in Fig. 5b.
Exposing CuO x /C to an equimolar amount of sulfur (M/S = 1) with a longer HTG reaction time (15 h) enabled Cu 1.8 S (D p = 34 nm) to be the predominant crystalline phase in the sample as outlined by XRD (purple curve in Fig. 5d). The Cu 2 O pattern is nearly invisible, and Cu(0) inexistent, indicating an extensive sulfidation of copper, in line with the high sulfidation rate (77%, see Table 3) in the spent material. After this long sulfidation test, TEM indicated a material composed of 20-100 nm NPs (majorly Cu 1.8 S) distributed within/on the C matrix, but also some large particle/agglomerate of a few hundred nanometers (see Fig. S9). Well-distributed < 5 nm nanocrystallites were also observed in the sample (Fig. S10), expected to be Cu 1.8 S and Cu 2 S according to lattice analysis. Considering their small size and low abundance, it is possible that these particles (at least the Cu 2 S particles) formed from a rapid reaction between dissolved Cu(H 2 O) x + and HS − upon quenching the reactor, a reaction that was shown to occur within a few minutes at room temperature 57 .
Overall, the results indicate a preferential sulfidation of Cu(0) over Cu 2 O, and the partial reduction of Cu(I) to Cu(0). Cu(0) sulfidation likely occurs by surface adsorption of H 2 S, or thiol followed by its subsequent decomposition and subsequent migration of S within already partially sintered Cu particles. Cu 2 O sulfidation seems to follow a dissolution/precipitation mechanism as suggested for ZnO. www.nature.com/scientificreports/ S-inert FeO x /C with high structural stability under HTG condition. FeO x /C presents high stability upon hydrothermal treatment with DEDS. No iron sulfide compounds could be detected by XRD. This is in contradiction with the absorption capacity results, indicating that the carbon matrix absorbed more S than expected from the absorption observed on the reference C sample. Comparing TEM (Fig. 6a,b) and PSD (Fig. S5c) results of spent and fresh FeO x /C, a limited impact of SCW and DEDS/isopropanol on the material (see Fig. 6c) was found, with a limited coarsening of FeO x particles growing from 15 ± 4 to 18 ± 5 nm, in phase with the limited crystallite growth observed by XRD (D p from 20 to 36 nm in average). Overall, the particle size distribution remains narrow and the particles are homogeneously distributed within the C matrix. The only crystalline phase in the fresh material and the material after the sulfidation test was Fe 3 O 4 , but in the latter, very weak diffraction peaks corresponding to Fe 2 O 3 were observed. These observations further confirm the poor ability of iron oxide to absorb sulfur. This is due to the lack of stability of iron sulfide in supercritical water previously reported 19,58 . This finding also discloses the high stability of the oxide with limited coarsening/sintering in SCW conditions. This property makes FeO x /C or Fe 3 O 4 suitable materials for other applications under SCW conditions, such as catalyst support or guard bed for downstream heterogeneous catalysts to retain e.g. transition metals or particles.
Evolution of MnO x /C upon sulfidation under HTG conditions. MnO x /C suffered from particle agglomeration, similar to ZnO/C and CuO x /C, after reacting with DEDS in the SCW environment. Big agglomerates are ubiquitous in the spent material as shown in Fig. 7b, with heterogeneous distribution of sulfur. The formation of MnS was confirmed by XRD shown in Fig. 7c. MnO is the major crystalline phase in fresh MnO x /C and after exposure to sulfur under SCW. The mean crystallite size D p decreased from 45 to 38 nm. The EDX results of a representative S-rich area suggests Mn, S, and O concentrations to be 26 mol%, 11 mol%, and 14 mol%, respectively. The sulfidation ratio of this area reaches 42%, showing a big difference compared to the bulk ratio of 9.3% (see Table 3). The heterogeneity of S-absorption has been also observed on porous ZnO/C and CuO x /C. It seems that the porosity of the MO x /C material has a very limited impact on the S-absorption homogeneity. www.nature.com/scientificreports/ CeO x /C with high structural stability upon sulfidation under HTG conditions. After the S-absorption reaction with DEDS under hydrothermal conditions, no large agglomerates are found in the spent CeO x /C, but some growth of the NPs from below 2-3 nm (Fig. 8a) to 5 ± 2 nm (Fig. 8b). The particle size distribution (see Fig. S5e) remains narrow and the particles are still homogeneously distributed within the C matrix. The average crystallite size of the main component CeO 2 increased from 4.1 to 5.7 nm with an increase of crystallinity, as indicated by XRD (Fig. 8d). Another very weak pattern is tentatively assigned to Ce 2 S 3 . EDX in Fig. 8c indicates a rather homogeneous distribution of S throughout the material, with a spatial distribution closer to that of C and O than Ce. A small proportion of S-rich areas are also rich in Ce, e.g. the top-left region, which should correspond to small highly-sulfided CeS x NPs. In summary of "Different structural evolution of MO x /C materials under HTG conditions" section, the above material characterizations disclose the significant morphological evolution of ZnO/C, CuO x /C, and MnO x /C by reacting with DEDS under HTG conditions. The distinct separation of MO x (M = Cu, Zn, Mn) from porous C matrix and the formation of big M-based agglomerates were observed. In contrast, FeO x /C and CeO x /C showed outstanding structural stability with MO x particle size remaining well-dispersed within the C matrix and below 36 and 6 nm, respectively. FeO x /C presented a limited capacity to absorb sulfur with no crystalline Fe sulfides generated. CeO x /C showed a good S-absorption capacity comparable to ZnO/C, but lower than CuO x /C, with the formation of small cerium sulfide NPs. S-absorption on the non-porous CeO x /C were homogeneous, in contrast with the porous ZnO/C and CuO x /C, which however showed heterogeneous S-absorption. It seems that the porosity of the MO x /C material has a very limited impact on the S-absorption homogeneity. A bold assumption is a faster rate of MO x /C structural evolution (MO x particle size growth and their migration away from the C matrix) than S-absorption under such conditions. Ostwald ripening, initiated by the dissolution of MO x particles, might be responsible for the migration of the Zn, Cu and Mn oxides away from the C matrix and the particle growth. In the next section, we use thermodynamic models to predict the dissolution of ZnO, Cu 2 O, Cu, Fe 3 O 4 , MnO, and CeO 2 in water under the conditions (T, P, ρ H2O ) applied in the experimental part of this paper to give further support to this assumption.
Thermodynamic modeling of MO x dissolution in water at different conditions. Reported solubility data of ZnO, Cu 2 O, Cu, Fe 3 O 4 , MnO, and CeO 2 under sub/supercritical water conditions are scarce in the literature. Based on the R-HKF method illustrated in "Thermodynamic modeling of MO x dissolution in water", the solubility of those metal oxides or metals in pure neutral water was calculated for eight different sets of conditions (T, P, ρ H2O ) along the heating-up process from liquid water (50 °C, 3.5 MPa, 0.99 g mL −1 ) to supercritical water (450 °C, 30 MPa, 0.14 g mL −1 ). The calculated solubility result is tabulated in Table S7 and shown  For all of those MO x materials, a higher solubility in hot liquid water (≤ 350 °C) than in supercritical water (≥ 400 °C) is observed (Fig. 9), while the lowest solubility is found at conditions where water is majorly in the vapor state (point at 375 °C and 217 bar) due to a rapid decrease of water's dielectric constant. Figure 9 indicates very limited solubility of Fe 3 O 4 and CeO 2 even in hot liquid water, with values lower than 10 -10 mol kg −1 and 10 -16 mol kg −1 , respectively. The solubility of ZnO, Cu 2 O, Cu, and MnO were found to be close to each other, and was calculated to be over five orders of magnitude higher than Fe 3 O 4 and nine orders of magnitude higher than CeO 2 . ZnO and Cu 2 O both show a dissolution increasing with a temperature rising from 50 to 300 °C, reaching at 300 °C their highest solubility of 10 -3.9 mol kg −1 and 10 -4.6 mol kg −1 , respectively. MnO is predicted to be preferentially dissolved in colder water and it reaches its highest solubility of 10 -3.4 mol kg −1 at 50 °C. The highest solubility of Cu(0) at the modeled conditions is predicted to be 10 -3.7 mol kg −1 at 200 °C.
These predictions agree with the experimental evidence of the structural evolution of MO x /C discussed in "Different structural evolution of MO x /C materials under HTG conditions" section. FeO x /C and CeO x /C showed outstanding structural stability with their MO x (M = Fe, Ce) NPs still well-dispersed within the C matrix after the HTG reaction, in line with the low solubility of Fe 3 O 4 and CeO 2 in water. In contrast, the migration of MO x (M = Zn, Cu, Mn) particles from the C matrix along with the growth of particle size observed on spent MO x /C (M = Zn, Cu, Mn) materials after HTG reaction correlate with the higher calculated solubility of ZnO, Cu/Cu 2 O, and MnO in sub/supercritical water. TEM analysis of spent materials brought evidence that Ostwald ripening is involved in the particle growth at least of Zn-and Cu-based particles ("Evolution of ZnO/C by SCW and upon sulfidation under HTG condition" and "Evolution of CuO x /C by SCW and upon sulfidation under HTG condition" sections). The correlation between higher particle growth and higher solubility supports the occurrence of Ostwald ripening. When it comes to tests in presence of a large amount of reduced sulfur compounds, the model used here does not apply anymore. Indeed, the acidification effect of dissolved H 2 S (decrease of pH) might promote the dissolution of MO x by a few orders of magnitude 59 . The correlation between the M-based particle growth and the heterogeneity of S-absorption was also observed by EDX. This suggests that MO x with different solubility have different S-absorption mechanisms. MO x with relatively high solubility, such as ZnO, Cu 2 O, and MnO, may absorb S by dissolution (to metal cations) and reprecipitation (forming metal sulfide species with lower solubility). In contrast, sulfidation of very low solubillity CeO 2 NPs may occur through surface adsorption of sulfur species (likely H 2 S or thiol) followed by S migration in the bulk of the material leading to the formation of cerium sulfide.

Conclusions
Metal oxide nanoparticles embedded in carbon, MO x /C (M = Cu, Ce, Zn, and Mn), have been shown to absorb sulfur at HTG conditions with different S-absorption capacities with the following order CuO x /C > CeO x /C ≈ ZnO/C > MnO x /C, through the formation of Cu 1.8 S, Ce 2 S 3 , ZnS, and MnS nanocrystals, respectively. However, the lack of stability observed for some active phases, such as Cu 2 O/Cu, ZnO, and MnO, could lead to a nonnegligible loss of performance and mechanical loss of material in a continuous flow process. Calculations showed that the solubility of the active phases is low, and thus they would not leach significantly during a continuous hydrothermal processing of wet biomass under supercritical water conditions. A correlation between MO x solubility and MO x /C structural evolution observed in the spent material, i.e. particle growth/migration of particle outside the C matrix, was found. Together with the evolution of particle size distribution, this indicates that Ostwald ripening through metal dissolution is likely the dominating cause for particle size growth. Interestingly, coalescence seems to be less favorable with sulfided metals than with the corresponding metal oxides, which is particularly visible for ZnO. Consequently, sulfidation of relatively more soluble metal oxides under HTG condition is believed to follow MO x dissolution/MS x precipitation mechanism, with strong evidence for ZnO, Cu 2 O, and MnO. The kinetics of nuclei migration in SCW, and the stability of the formed metal sulfide particles in long-term operation remains to be determined. Fe 3 O 4 was found to have a high stability with very low capacity to absorb sulfides during the HTG process. These properties make this material an excellent candidate as a long-lasting safeguard bed to protect downstream catalysts from poisoning by other transition metals or solids, or as a stable support in composite S-absorbents. CeO x /C exhibited a high structural stability and good S-absorption capacity, and is suggested as a promising sulfur absorbent in such applications. Further efforts should be devoted towards developing the synthesis of CeO x /C to achieve high CeO x loading with high porosity, as well as efforts in assessing the regeneration of the sulfided material.

Data availability
The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.