Construction of anhydrous two-step organosolv pretreatment of lignocellulosic biomass for efficient lignin membrane-extraction and solvent recovery

Glycerol organosolv (GO) pretreatment has been revealed to be potent in selectively deconstructing the lignocellulosic biomass and effectively enhancing its enzymatic hydrolysis, but the conventional solid washing and GO lignin extraction processes frequently consume large amounts of water, resulting additionally in difficulty recycling the glycerol. In this study, an anhydrous two-step organosolv pretreatment process was explored, followed by the membrane ultrafiltration of glycerol lignin. The results showed that the solid washing of the residual glycerol after the atmospheric glycerol organosolv (AGO) pretreatment was necessary for the subsequent operation of high-solid enzymatic hydrolysis. Washing with ethanol was desirable as an alternative to water as only a low glycerol content of 5.2% resided in the substrate. Membrane ultrafiltration was helpful in extracting the AGO lignin from the pretreatment liquor, in which a high lignin extraction of 81.5% was made with a regenerated cellulose membrane (cut-off for 1 kDa) under selected ultrafiltration conditions. With the characterization of membrane-extracted lignin, it was observed for the first time that the AGO lignin has a well-preserved structure of G/S type. Moreover, the lignin was enriched with reactive groups, i.e. β-O-4′ linkages and aliphatic hydroxyl groups, which was very likely due to the glycerol grafting onto the lignin via α-etherification reaction. The two-step organosolv pretreatment process allowed 86% of glycerol and 92% of the ethanol recovery with ∼78% of distillation energy savings, which was applicable for extraction of organosolv lignin and recycling use of organic solvents.


Introduction
The emergence of several environmental threats, together with the depletion of fossil fuels, has attracted extensive attention around the world in recent years to the production of renewable energy from lignocellulosic biomass that is environmentally friendly and carbon-neutral [1]. Notably, the inborn architecture of lignocellulosic biomass, mainly consisting of cellulose, hemicellulose, and lignin, forms a rigid shield that restrains it from external environmental damage [2]. Accordingly, pretreatment has been regarded as a crucial step in the enzyme-mediated lignocellulosic biorefinery as it is deemed to deconstruct the lignocellulosic biomass into components that are accessible to cellulase enzymes [3][4][5][6][7]. Thus, the pretreatment technique can decisively influence downstream processes of enzymatic hydrolysis and microbial fermentation, as well as by-product valorization.
Among dozens of methods, organosolv pretreatment is one of the most deeply studied techniques [8][9][10][11]. The organosolv pretreatment of lignocellulosic biomass has presented some distinct advantages, i.e. great component selectivity, reduced cellulose crystallinity, and considerable delignification. Remarkably, lignin isolated from the organosolv process, namely organosolv lignin, has a narrow molecular weight distribution and higher purity, together with abundant reactive groups, thus rendering it extremely suitable to prepare macro-molecular phenolic polymeric materials [12][13][14]. Conventional organosolv pretreatment refers to treatment with low-boiling-point organic solvents, i.e. methanol/ethanol, formic/acetic acid, and acetone, as these solvents are easy to recycle after cooking. However, the low-boiling-point organosolv pretreatment has always been limited in the laboratory, owing to the risky operation, i.e. volatile, combustible, explosive, and sometimes toxic, in addition to the high pressure [15].
Notably, glycerol is a very safe organic solvent with a high boiling point, which is also a main by-product of the oleochemical industry. Sun et al [12,[16][17][18] initially developed an auto-catalytic atmospheric glycerol organosolv (AGO) pretreatment process ten years ago, in which the process displayed an expectedly high selectivity of components and afforded a remarkable enzymatic hydrolyzability of wheat straw. Moreover, many other researchers have testified to it by developing the glycerol organosolv (GO) pretreatment process for the fractionation of various lignocellulosic biomass [8,[19][20][21][22][23]. For example, sugarcane bagasse pretreated with sodium methoxide and glycerol displayed a high cellulose preservation of ∼89% and a high lignin removal of ∼80% [24]. The GO pretreatment of bagasse at 198 • C for 1.5 h obtained 54.4% of pulp yield, 13.7% of polyose content and 81.4% of delignification [25]. More encouragingly, these GO pretreated substrates have also displayed outstanding hydrolyzability of the biomass. Hassanpour et al [13] showed that the hydrolysis yield of glucan from sugarcane bagasse was as high as 99% after the acid-catalyzed GO pretreatment. Gabhane et al [20] observed that the GO pretreatment of rice straw led to a 71.5% reduction in sugar yield with 94.3% holocellulose digestibility. In brief, the GO pretreatment presented industrial attractiveness with excellent component selectivity and boosted substrate hydrolyzability. Nevertheless, most of these studies on GO pretreatment have concentrated on the undissolved solid fiber fractions for the release of fermentable sugars, whereas there is scarce research work on the pretreatment liquor, i.e. the lignin extraction and its structural features.
Generally, lignin extraction is performed in the laboratory with differential precipitation, in which the lignin is recovered by adjusting the pH value of the solution or adding an excess of water [26]. Lourençon et al [27] separated the lignin from hardwood and softwood Kraft black liquor by acid precipitation that was carried out sequentially by acidification at different pH (9, 7, 5, 3, and 1) values, and then evaluated the differences in these separated lignins. Lan et al [28] precipitated the lignin from the pretreatment solution of sugarcane bagasse pretreated with p-toluenesulphonic acid by using ten-fold deionized water. Likewise, Smink et al [29] studied the extensive addition of water as an anti-solvent for lignin precipitation during lignocellulosic component fractionation using deep eutectic solvents (DES). Nevertheless, it is very energy-intensive to recover the DES as the high amount of water addition excessively dilutes the solvent and the anti-solvent must be evaporated from the DES. In the case of the GO pretreatment process, the excessive use of water in the GO pretreatment liquor to obtain lignin is far from industrial interest, as it not only leads to large water consumption and environmental damage from the effluent, but also discourages the recycling use of glycerol. Additionally, the use of water in the pretreatment process dilutes the glycerol content, which results in a high energy consumption in concentrating the glycerol for recycling use by water evaporation. Therefore, a new GO pretreatment process that typically affords efficient lignin extraction with little or no water use is desirable for the development of an enzyme-mediated lignocellulosic biorefinery. Alternatively, it is favorable to seek a low-boiling-point solvent that can contribute to lignin fractionation as well as low energy recovery with replacement of water.
In this study, an AGO pretreatment process coupled with ethanol washing and membrane ultrafiltration was attempted for lignin extraction and solvent recovery. First, the effect of the glycerol residing in the lignocellulosic substrate after AGO pretreatment on enzymatic hydrolyzability was evaluated to ensure whether the solid-washing process was required. Then, ethanol washing of the solid was used as an alternative to water, followed by construction of the ethanol-washing process. To extract AGO lignin, a membrane ultrafiltration process was explored with a selection of the main variables. Finally, the extracted AGO lignin from the pretreatment liquor was characterized.

Materials
Sugarcane bagasse was harvested in Guangxi Province, China. It was sieved through a 20-mesh sieve (0.83 mm) and dried at 105 • C for 8 h to maintain a constant weight, in accordance with previous reports [30]. The main components of sugarcane bagasse contained 41.0% cellulose, 24.3% hemicellulose, and 23.7% lignin. The cellulase preparation Cellic CTec2 (120 FPU g −1 ) was generously gifted by Novozymes Investment Co. Ltd, Beijing, China. Industrial glycerol (⩾99.5% purity) was provided by a chemical plant in Jiangsu Province, China. Other chemicals were purchased from Sinopharm Chemical Reagent Company Ltd, China.

AGO pretreatment process
In a typical run, 10 g of dried sugarcane bagasse was placed in a triple-necked round-bottom flask that contained 140 g (solid/liquid ratio of 1/14, w/w) of glycerol, and 2.8% (w/w) NaOH based on the feedstock was mixed into the flask. Then, the flask was heated at 240 • C for 20 min. The pretreatment conditions were selected from our previously reported work [30,31]. Undissolved solids and pretreatment liquor fractions were separated through a G1 glass filter. The pretreatment liquor fractions were collected and kept at 4 • C to use in the subsequent membrane ultrafiltration. Thereafter, the solid washing process was conducted using ethanol as the detergent, and washing with water was or was not performed as the control groups. Different volumes of solution for washing (based on the ratio of raw substrate to medium amount of 1-50 w/v) were mixed with the undissolved solid in a beaker, followed by stirring for 5 min, and then the mixture was separated by filtration. The undissolved solids were collected for analysis of biomass composition, and they were also used to assay substrate hydrolyzability in the enzymatic hydrolysis process.
The solid recovery rate is defined based on equation (1), Solid recovery rate (%) = Weight of pretreated solid (g) Raw substrate weight before pretreatment (g) × 100.
To estimate the content of glycerol retained on the surface of the pretreated substrate, 1 g of each of the washed or non-washed samples was mixed with 20 ml of deionized water in a 50 ml centrifuge tube. After mixing upside down for 3 min, it was centrifuged (8000 rpm) for 10 min at room temperature. The supernatant was collected for assaying the glycerol content.

Enzymatic hydrolysis 2.3.1. Enzymatic hydrolyzability of different washing substrates
The non-washed and washed undissolved solid fractions were used for enzymatic hydrolysis with a solid concentration of 5% (w/v), i.e. 1.25 g of dry weight equivalent. The enzyme hydrolysis was made at 5 FPU g −1 substrate by cellulase loading with 25 ml of citrate buffer (0.05 M, pH 4.8), followed by incubation at 50 • C and 180 rpm for 72 h. At the end of the hydrolysis process, a 0.4 ml sample was withdrawn and heated at 100 • C for 5 min to inactivate the enzyme, then centrifuged at 10 000 rpm for 5 min. The supernatant was used to test the glucose concentration.

Assay of the inhibitory effects of glycerol on enzymatic hydrolysis
The enzymatic hydrolysis of fully washed, pretreated substrates and microcrystalline cellulose used as a model substrate was carried out as above. To evaluate the role of glycerol, different glycerol contents (0.0%, 0.25%, 0.5%, 1.0%, 2.0%, 3.0%, 5.0%, 7.0%, and 10.0%, w/v) were added in the enzymatic hydrolysis. In another trial, different glycerol additions (1%, 2%, 3% and 5%) in the enzymatic hydrolysis at various cellulase loadings (5.0, 7.5, and 10.0 FPU g −1 ) were also made for further evaluation. An inhibition coefficient was introduced to analyze the inhibitory role of glycerol, as shown in equation (2), where T 1 and T 2 represent the glucose concentration (g/l) released from the enzymatic hydrolysis in the absence or addition of glycerol, respectively.

Ultrafiltration separation process 2.4.1. Ultrafiltration trials
The pretreatment liquor was micro-filtered using an organic membrane of 0.45 µm to remove a large lignin fraction prior to ultrafiltration. Cross-flow ultrafiltration was carried out at room temperature using a regenerated cellulose membrane (P2C01MC01, MERCK MILLIPORE) with 1 kDa of the molecular weight cut-off and 0.1 m 2 of the total surface area. The pure water flux of the cellulose membrane was measured at an ultrafiltration pressure of 0.2 MPa and a flow rate of 190 ml min −1 . After obtaining a stable water flux, the permeated volume of water was recorded at a certain time interval. Additionally, around 150 g of pretreatment black liquid containing various concentrations of glycerol and ethanol were run for 1 h at a certain operating pressure and flow rate conditions, and the lignin content of the permeated and intercepted liquor was determined. The basic operating conditions were set as 0. After each round of lignin recovery, this membrane was reused for the next batch experiment using a simple backwash and a 0.1 M sodium hydroxide rinse. The flux calculation formula is shown in equation (3), where J represents the membrane flux, and V, S, and t refer to the volume of infiltration capacity (l), membrane area (m 2 ), and ultrafiltration time (h), respectively.

Determination of lignin interception rate
To quantify the lignin interception rate during ultrafiltration, the initial organosolv lignin concentration in the mixture of pretreatment liquid and ethanol was first determined by a spectrophotometry method using the purified lignin as a standard model. The purified lignin could be obtained in the following steps: (a) the pretreatment liquid was mixed with excessive acid water (adjusting deionized water to pH 2 with 6 M HCl) to precipitate the lignin samples; (b) the lignin samples were dissolved in 90% (v/v) acetic acid at a solid-liquid ratio of 1/20 (g ml −1 ), and the acetic acid-lignin mixture was added dropwise to ten times the volume of acid water; (c) the supernatant was removed by centrifugation, and the precipitate was washed three times with acid water and freeze-dried after centrifugation to produce the purified lignin. This lignin is named acid-precipitated lignin. The absorbance of the purified lignin dissolved in ethanol solution was measured at 280 nm, and the established relationships between the absorbance and lignin solution concentration were used to calculate the organosolv lignin concentration. Based on these, the lignin interception rate (R L , %) was determined by equation (4), where C 0 and C 1 refer to the initial and permeated lignin concentration (g/l), respectively.

Recovery of lignin, glycerol, and ethanol
The recovered lignin consisted regularly of two fractions: (a) a lignin cake layer remaining on the 0.45 µm microfiltration membrane, and (b) a concentrated lignin liquor intercepted by the ultrafiltration membrane. For the first fraction, the lignin cake was washed twice with 100 ml of acid water, then centrifuged and air-dried to recover the lignin. For the other fraction, the concentrated liquor was dropped into ten times the volume of acid water to form lignin precipitation, and the precipitated lignin was also obtained after centrifugation and air-drying. The lignin recovery rate was calculated according to equation (5) as follows: M 0 is the total mass (g) of the extracted lignin; M 1 and W 1 represent the mass (g) of raw material and its lignin content (%), respectively; and M 2 and W 2 represent the mass (g) of the undissolved solid after the pretreatment and its lignin content (%), respectively.
After the membrane ultrafiltration, ethanol in the permeate liquid was recovered at 50 • C and 0.1 MPa using a rotary evaporator. The recovery of ethanol was calculated based on the percentage of recovered ethanol volume to the actual volume of consumed ethanol in the sample. Similarly, the recovery of glycerol was calculated from the percentage of the remaining glycerol weight after evaporation to the actual weight of glycerol used in the sample.

Analytical methods
The glycerol lignin obtained from the ultrafiltration separation was characterized for its physicochemical properties. Thermogravimetric analysis (TGA) was carried out by using a thermal analyzer (Q500, TA Instruments, USA) in the temperature range of 30 • C-600 • C with a heating rate of 10 • C min −1 and a nitrogen flow rate of 30 ml min −1 [12]. Fourier transform infrared (FT-IR) spectroscopy was performed on a NICOLET NEXUS 470 equipment (Thermo Nicolet, America) in a KBr tablet in the wavenumber range of 400-4000 cm −1 , and the number of scans was 128 [32]. The 2D-HSQC nuclear magnetic resonance (NMR) was conducted based on the previous literature [33].
The main components (i.e. cellulose, hemicellulose, and lignin) of lignocellulosic samples were analyzed using a two-step acid hydrolysis method (National Renewable Energy Laboratory, NREL). A high-performance liquid chromatograph (Chromaster CM5110, Hitachi, Japan) equipped with an Aminex HPX-87H column (9 µm, 300 × 7.8 mm, BioRad, USA) and a refractive index detector were used to detect the concentrations of glucose, glycerol and ethanol [12]. The glucose yields were calculated by equation (6),

Effect of glycerol residing in the substrate on subsequent enzymatic hydrolysis
Our previous research detected extremely low toxic by-products (i.e. <1g kg −1 of furfural and 5-HMF) generated during the AGO pretreatment [17]. In this case, the main inhibition of enzymatic hydrolysis of AGO pretreated substrates may be caused by the large glycerol residual in the substrate, thus meaning that the substrate washing process should be necessary. To verify such an assumption, the influence of the glycerol on enzymatic hydrolysis was first studied (figure 1). When 0.25%-0.5% (w/v) of glycerol was added into the substrate, there were no remarkable effects on enzymatic hydrolysis ( figure 1(a)). With a further addition to 1.0%-10%, the glucose yield decreased obviously from 64.0% to 47.2%. Likewise, the same presence of glycerol during enzymatic hydrolysis of microcrystalline cellulose exhibited an even greater negative impact, resulting in the glucose yield decreasing sharply from 48.3% to 25.7% ( figure 1(b)). Theoretically, the lowered glucose yield from enzymatic hydrolysis presents a significant linear correlation with glycerol addition (figure S1), indicating that the excess glycerol residing in the substrates was indeed capable of hindering the hydrolysis process. This result is in accordance with previous studies [22,34]. The absorbed glycerol likely increased the viscosity of biomass slurry, and thus hindered the enzyme attack on the substrates [22]. It is possible that glycerol acting as a glucose analog led to an end-product inhibition of cellulase activity [34]. Additionally, the obvious decline in the glucose yield (4.2%) started at 1% of glycerol addition in enzymatic hydrolysis. The threshold concentration of glycerol is obviously lower than that reported at 5% by Zhang et al [34] and 2% by Zhu et al [22], which may be attributed to the different enzyme loadings on the substrate. To verify this, we studied the effect of glycerol concentration (1%-5%) on enzymatic hydrolysis of substrates under different enzyme loadings (5-10 FPU g −1 substrate) (figure S2). Interestingly, there was no significant inhibition within 1%-2% of glycerol concentrations when the enzyme loading increased to 10 FPU g −1 substrate, and only 4% of inhibition coefficient was observed at 3% of glycerol concentration in the entire enzymatic hydrolysis. The results indicate that the cellulase loading on the substrate should influence the threshold concentration of glycerol addition. The threshold concentration of glycerol addition was evidently low in the enzymatic hydrolysis of a substrate with low cellulase loading. Considering the threshold glycerol concentration of 1%, it was estimated that the residual glycerol should be controlled within 9% based on the substrate when the substrate is taken for subsequent high-solid hydrolysis (20% solid content) with 5 FPU g −1 of cellulase dosage. Thus, it was reasonable to remove the excess glycerol from the substrate by solid washing in the AGO pretreatment process.

Solid washing process after cooking
Experiments of washing the undissolved solid with tap water and ethanol were made after cooking. As shown in table 1, the AGO pretreated substrate contained 5.4% and 3.8% of glycerol residual, respectively after washing with water and ethanol, far lower than that (123.5%) of the non-washed pretreated substrate. As for the main components of substrate, the ethanol washing resulted in 92.0% cellulose retention and 78.3% hemicellulose retention, and 81.5% of delignification, which is obviously higher than that of water washing.
The data indicate that the ethanol washing in the pretreatment process was more helpful for the selective fractionation of the main components of undissolved solids after cooking. Moreover, the low boiling point of the ethanol is helpful to save energy in the subsequent ethanol recovery process by distillation, so it is desirable to select ethanol washing of the undissolved solid after cooking.
After the undissolved solid washing, glucose yield was used to evaluate the enzymatic hydrolyzability of the washed substrate (72 h, 5 FPU g −1 substrate, 5% substrate, w/v). The glucose yield from the ethanol-and water-washed substrates was 62.5% and 67.1%, which is almost 20% and 30% higher, respectively, than that  of non-washed. The significant enzymatic hydrolyzability is very likely due to the extremely low presence (0.27% and 0.21%, respectively) of glycerol in the enzymatic hydrolysis of both water-and ethanol-washed substrates and the excess presence (5.7%) of glycerol in that of the non-washed substrate. The results indicate that the undissolved solid washing process was a vital step in improving the component fractionation and subsequent substrate hydrolyzability. Notably, the enzymatic hydrolyzability of the ethanol-washed substrate was slightly lower (∼7%) than that of the water-washed, which was because the exposed native lignin after the ethanol washing removed the extractable lignin deposited on the substrate surface should have presented stronger adsorption to cellulase enzymes [35,36]. As reported previously, the lignin residing in the undissolved solid fraction mainly contained two sections, namely the intact native lignin embedded in the interior and the solvent-extractable lignin deposited frequently on the outer surface [35,36]. The former lignin was more depolymerized and less condensed, thus resulting in lower ineffective adsorption of the cellulase enzymes. Recently, the removal of these extractable lignin fragments from the substrate by ethanol washing was surprisingly observed to have an adverse effect on the enzymatic digestibility of organosolv-pretreated sweetgum [37]. In short, the extractable lignin fragments deposited on the substrate surface alleviated the occurrence of enzyme non-productive binding by sheltering the enzyme binding sites on the residual bulk lignin. To improve the enzymatic hydrolysis of the ethanol-washed substrate, a less expensive additive, Tween 80, was used [38]. As shown in figure 2, the glucose yield from the ethanol-washed substrate increased similarly to that of the water-washed one with the greater Tween 80 addition. At 40 mg g −1 of Tween 80 addition, the glucose yield of these two substrates was almost equivalent (∼76%), indicating that the positive role of Tween 80 was more significant for enzymatic hydrolysis of the ethanol-washed substrate with more exposed native lignin [39,40]. The data suggest that the ethanol-washed pretreated substrates expose more residual native lignin and lead to stronger non-productive adsorption of cellulase. The addition of Tween 80 was extremely effective in reducing the non-productive adsorption, which was responsible for the highly enhanced glucose yield. This is in accordance with another study in which the addition of bovine serum albumin obviously increased the glucose yield of the ethanol-washed acid-pretreated larch [35]. Consequently, it is desirable to select ethanol washing of the undissolved solid fraction after cooking in the AGO pretreatment process.

Construction of the ethanol-washing process
The ensuing experiment was to select reasonable ethanol-washing volume and washing time for the ethanol-washing process after cooking ( figure 3). As shown in figure 3(a), the percentage of glycerol residing in the substrates lowered gradually with a greater volume of ethanol washing, and relatively the enzymatic hydrolysis of the washed substrate improved obviously. At 300 ml of ethanol washing volume, the glycerol remained at 23.6% based on the washed substrate, and the glucose yield reached 62.5% from it. Thereafter, the enzymatic hydrolysis increased very slightly with more ethanol washing volume. Thus, 300 ml of ethanol washing volume was selected for the next experiment. Then, the ethanol washing time was investigated ( figure 3(b)). Comparing the immediate ethanol washing with 300 ml, all of the multi-time ethanol washings resulted in a sharp glycerol descent from above 20% to less than 10% residing in the substrate, despite a slight increase in enzymatic hydrolysis. Among them, the mode of ethanol washing twice with 150 ml of ethanol was favorable as the residual glycerol in substrates was as low as 5.2%, rendering a high glucose yield of 65.9% after 72 h of enzymatic hydrolysis. Accordingly, the solid washing process was constructed with 150 ml of ethanol washing twice.
Additionally, recent studies have demonstrated that glycerol should participate in the chemical reaction with by glycerolysis [13,14]. Glycerol is a typical polar molecule containing multiple hydroxyl groups. During the high-temperature glycerol pretreatment process, glycerol participates in glycosyl modifications and reactions with lignin in pretreatment liquor. The dissolved sugars dissociated from lignocellulosic biomass present in the pretreatment liquor can be combined with glycerol in the form of glyceryl glucoside/xyloside [41]. Zhang et al [41] discovered that when the liquor was diluted with distilled water and then hydrolyzed, the xylose content increased from 8.5% to 62% of the xylan in the hydrolysates, along with the detection of glucose (1.3%). This implied that glucose/xylose was released from glycerol glucosides/xylosides. Conceptually, all these sugars that detach from biomass are more likely to form glycerol glycosides with glycerol, since glycerol can react with the -OH groups of xylose/glucose and oligosaccharides during GO pretreatment. In addition to glycerol glycosides, several studies have reported glycosides produced by organosolv reaction, namely ethyl xyloside/glycoside [42][43][44] and alkyl xyloside/mannoside [45]. Accordingly, this organosolv glycosylation reaction ought to be used extensively in different organosolv pretreatment processes.
Apart from modifying the glycosyl group, glycerol may also be chemically bound to the pretreated solid component. During the high-temperature glycerol pretreatment process, part of the glycerol is attached to the substrate surface in a free form (this glycerol part can be easily washed away), and the other part may be bound to the insoluble solids in a chemical bonding state such as a hydrogen bond or covalent bond (this glycerol part can be difficult to wash off). Therefore, a reasonable washing process is not only beneficial to save detergent consumption, but also required for the development of enzyme-mediated lignocellulosic biorefinery.

Isolation of GO lignin by ultrafiltration
Several studies have been conducted on membrane concentration or recovery of by-products from black liquors using ultrafiltration [46,47]. In this study, a regenerated cellulose membrane was used for the ultrafiltration separation of lignin. The effect of the operating conditions (solvent concentration, pressure, flow rate, and temperature) on membrane flux and lignin interception rate was evaluated (figure 4). As shown in figure 4(a), the membrane flux decreased from 1.87 to 1.2 l (m 2 h) -1 with the glycerol concentration increasing from 35% to 50%. The decrease in membrane flux [0.69 l (m 2 h) -1 ] was more apparent when the glycerol concentration reached up to 55%. Based on this, the glycerol concentration was selected as 50% in the next experiment.
As shown in figure 4(b), the membrane flux gradually increased from 0.17 to 0.61 l (m 2 h) -1 with an increase in operating pressure from 0.1 to 0.2 MPa, meaning that the high operating pressure provided a strong driving force to penetrate glycerol through the membrane. As can be seen in figure 4(c), the membrane flux increased to 0.68 l (m 2 h) -1 from 0.57 l (m 2 h) -1 with the flow rate increasing from 130 to 190 ml min −1 . Increasing the flow rate probably reduced the concentration polarization on the membrane surface and the pore blockage, thus increasing the membrane flux [48]. In terms of the ultrafiltration temperature figure 4(d), the flux continually increased from 0.61 to 0.96 l (m 2 h) -1 when the temperature was increased from 25 • C to 45 • C. Additionally, the viscosity of glycerol was closely associated with the temperature, i.e. 0.01 Pa s at 110 • C from 1.4 Pa s at 20 • C. This increase was very likely due to the reduced viscosity of glycerol as the high temperature caused the liquor to flow more easily, allowing small molecules (i.e. glycerol and ethanol) to pass through the ultrafiltration membrane more easily [49]. Considering the temperature limit of the regenerated cellulose membrane at 50 • C, the operating temperature was selected at 45 • C. In addition, the interception rate of lignin was constantly stable at above 90% despite the changes in operation conditions including solvent concentration, pressure, flow rate, and temperature, suggesting that the lignin recovery depended less on the operating conditions. Under these selected operating conditions during ultrafiltration (50% of glycerol concentration, 0.2 MPa of pressure, 190 ml min −1 of flow rate, and 45 • C of temperature), the lignin was extracted at 81.5%, together with 86% glycerol recovery and 92% ethanol recovery.
Furthermore, the overall flow chart describing the AGO pretreatment process with ethanol washing followed by membrane ultrafiltration to extract lignin and solvent recovery is shown in figure 5. The ethanol usage for solid washing and subsequent recovery was also evaluated by comparison with water washing. The ethanol usage was less than 30% than the water consumption when they washed the solid till up to the same low glycerol content (5.4% vs 5.2%), indicating a stronger performance in glycerol removal from the undissolved solid substrate. Intriguingly, ethanol can be efficiently recovered through the conventional distillation process with even lower energy consumption, consuming <900 kJ kg −1 ethanol evaporation, far lower than the water evaporation (2260 kJ kg −1 ) under standard atmospheric pressure [50]. Given the assumption of 90% distillation efficiency in this study, the total heating energy required to evaporate ethanol was 206.7 kJ, equivalent to a nearly 78% saving in energy consumption compared with water distillation (924.1 kJ). Overall, the use of ethanol in washing the solid in the AGO pretreatment process is desirable concerning energy savings in solvent recovery. The recovered glycerol and ethanol could be further utilized for the next batch of pretreatment and subsequent ethanol solid washing. Lignin extraction of 81.5% was  also obtained after membrane ultrafiltration. Actually, the reagent-free membrane treatment allows the extraction of lignin fractions with a defined molecular weight distribution, rendering it more economically competitive by generating new revenue streams and reducing downstream costs [51]. However, there is scarce information on the application of membrane separation in organosolv liquor, and even less on the extraction of GO lignin.

Characterization of membrane-isolated AGO lignin
The physicochemical properties of the lignin separated by ultrafiltration were analyzed based on various structural characterizations. Figure 6(a) shows the TGA of the AGO lignin separated by membrane ultrafiltration and acid precipitation, which can be used to identify the polymerization degree and chemical bond strength of retrieved lignin polymers. The thermal weight losses of these two samples exhibited a slight degradation at 25 • C-200 • C, owing to the evaporation of water retained in the lignin fragments, and an obvious degradation at 200 • C-300 • C, which was mainly attributed to a loss of lignin with weak bonds and small molecular weight [52]. A rapid thermal degradation stage was observed in the temperature range 300  [53], suggesting rich β-O-4 ′ linkages and side chains contained in these AGO lignin fractions. Moreover, it is evident that the AGO lignin separated from membrane ultrafiltration contained relatively small amounts of β-O-4 ′ bonds compared with the acid-precipitated lignin. The thermal degradation rate of the former lignin was also slower than that of the latter, indicating that the membrane-ultrafiltered lignin should have relatively low side-chain units. Above 400 • C, the methoxy groups, aromatic rings, and lignin internal linkages underwent severe fracture. Briefly, the AGO lignin was likely enriched with reactive groups such as β-O-4 ′ linkages and side chains, and the favorable thermal stability allowed the AGO lignin to be used for preparation of lignin-derived materials in a broad temperature range, as well as undoubtedly for chemical transformations.
The lignin samples were validated by the FT-IR spectra as shown in figure 6(b). Some typical bands at around 1600, 1511 and 1425 cm −1 were assigned to the skeletal stretching vibrations of aromatic ring structures in lignin, and deformation vibration of C-H combined with aromatic rings was observed at 1460 cm −1 [54]. These vibrations were observed in both samples, indicating the extracted lignin products. The band at around 1328 cm −1 was related to the syringyl and condensed guaiacyl rings [55]. Meanwhile, the syringyl aromatic C-H in-plane deformations (1234 and 1126 cm −1 ), guaiacyl aromatic C-H in-plane deformation (1032 cm −1 ), and aromatic C-H out-of-plane in positions 2 and 6 of syringyl (835 cm −1 ) were determined by the FT-IR spectra [55]. Based on this, most of the isolated samples were typical for the guaiacyl (G) and syringyl (S) types of lignin. Due to the absence of the band at 1165 cm −1 (stretching vibration of p-hydroxyphenyl C=O) [56], it was inferred that the p-hydroxyphenyl (H) type of lignin degrades more efficiently than the G/S type during AGO pretreatment. Additionally, no other obvious signals in either AGO lignin sample were detected in the bands around 1734-1702 and 1123-1008 cm −1 that are assigned to the non-conjugated carbonyl-carboxyl stretching vibration from ketones, carbonyls and ester groups, as well as the complex vibrations with the C-O stretching, C-C stretching, and C-OH bending in polysaccharides [57]. The data suggest that the AGO pretreatment efficiently dissociated the lignin-carbohydrate crosslinks, resulting in a high purity of AGO lignin. More meaningfully, the samples of the lignin separated by membrane ultrafiltration and acid precipitation only exhibited slight changes in the structural segments and functional groups. Owing to the limited effects of interception on the aromatic rings, a core structure of lignin was maintained that was suitable for further conversion to produce aromatic chemicals.
These lignin fractions were further characterized with 2D HSQC NMR spectra (figure 7 and  [13,14], the alkaline-catalyzed GO lignin had no obvious signals of B α (δ C /δ H 86-90/5.4-5.8) and B γ (δ C /δ H 58-60/3.5-4.1) that are mainly assigned to the β-β linkages in the lignin. This suggested that the latter had a relatively higher content of β-O-4 ′ bonds while containing fewer β-β bonds. More interestingly, compared with other pretreatment methods such as dilute acid [13,14], some new signals (marked as M) at δ C /δ H 69.3-75.3/3.5-4.25 ppm were observed in the spectra, and their strength was significantly higher than that from the acid-catalyzed GO lignin. It was assumed that these signals were related to the etherification of glycerol with the α-position of β-O-4 ′ linkages. This observation is in line with previous arguments that the hydroxyl group on the α-position of β-O-4 ′ linkages would be protonated via the S N 1 nucleophilic substitution reaction under an acidic environment during the alcoholic organosolv (including glycerol) pre-treatment, thereby forming a reactive benzylic carbocation intermediate [13,45,58]. Subsequently, the hydroxyl groups in the alcohols likely reacted with the intermediate as a nucleophile and finally produced α-etherified lignin. A similar phenomenon happening in the side-chain region could occur extensively in various alcohol-based lignocellulose pretreatments, i.e. methanol, ethanol, n-butanol, and 1,4-butanediol [23,59]. Additionally, new signals at δ C /δ H 62.6/3.35, 65.3/4.05, and 69.3/3.70 ppm corresponded to an esterification reaction of glycerol with the γ-position of ferulic acid and p-coumaric acid, thus presumably generating feruloyl glycerol and coumaroyl glycerol [13]. Briefly, the AGO pretreatment tended to the glyceryl lignin phenolics through chemical modification with the grafting of glycerol, which avoided the excessive condensation of lignin fragments. The glycerol-grafted modification rendered the AGO lignin aromatic monomers enriched with the high percentages of β-O-4 ′ linkages and aliphatic hydroxyl groups, which was extremely beneficial for the valorization of AGO lignin. It can also be presumed that such grafting reaction should be more sensitive in the presence of an alkaline catalyst, as alkali was more favorable for delignification compared with an acidic catalyst.
In the aromatic region (δ C /δ H 100-140/5.5-7.8), the signals corresponding to the aromatic rings of syringyl (S), guaiacyl (G), and p-hydroxyphenyl (H) units were distinctly detected at δ C /δ H 103.6/6.73 (S 2,6 ), 110.9/6.98 (G 2 ), 115.6/6.77 (G 5 ), 119.0/6.79 (G 6 ), and 127.2/7.01 (H 2,6 ) ppm, respectively. It should be noted that the signals assigned to the H unit were observed at relatively lower intensity, which is consistent with previous studies on wheat straw, sugarcane bagasse, and corn stover [14,60]. This result suggests that the AGO pretreatment preferentially decomposed H units of lignin over G/S units. Further, there was no obvious difference in the main structures between the acid-precipitated lignin and membrane-extracted lignin. Nevertheless, the relative intensities of the G/S/H units from membrane-extracted lignin decreased to a certain extent, and evidently, the lignin substructures of PCA 2,6 (δ C /δ H 130.0/7.5), FA 2 (δ C /δ H 111.1/7.32), and F β (δ C /δ H 125.5/6.3) almost disappeared compared to the former. Therefore, the membrane-extracted lignin likely had relatively low amounts of these components, simultaneously giving the lignin a low molecular weight and a narrow molecular weight distribution.

Conclusions
An anhydrous two-step organosolv pretreatment of lignocellulosic biomass was successfully constructed for efficient lignin membrane separation and downstream solvent recovery. Too much glycerol residing in the undissolved solid fraction during the AGO pretreatment process was averse to the subsequent enzymatic hydrolysis. Ethanol washing of undissolved solids in this process is extremely desirable for lignin extraction and glycerol recovery. The membrane ultrafiltration recovered 81.5% of the dissolved lignin from the pretreatment liquor, and the glycerol recovery process allowed for 86% of glycerol and 92% of ethanol with 78% of distillation energy savings. The extracted lignin was enriched with abundant reactive groups, i.e. high percentages of β-O-4 ′ linkages and aliphatic hydroxyl groups, which was likely due to glycerol grafting onto the lignin via α-etherification reaction. The anhydrous two-step organosolv pretreatment containing ultrafiltration is of innovative promise for extraction of organosolv lignin and recycling use of organic solvents in the development of enzyme-mediated lignocellulosic biorefinery.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.