A global approach to obtain biobutanol from corn stover

Abstract The aim of this research was to subject corn stover to a complete biorefinery process at laboratory-scale in order to assess the production of biobutanol. The research was conducted to focus on process simplification, reduction of reagents and optimization of acetone-butanol-ethanol (ABE) fermentation. The main recommendations include the use of low acid concentrations during the physicochemical pretreatment, the selection of adequate Clostridium strains, detoxification of the hydrolysates with reusable adsorption resins and the possibility of performing gas stripping offline to recover ABE solvents. Various pretreatment conditions, fifteen bacterial strains and three polymeric adsorption resins were assessed. The proposed method consisted of a physicochemical pretreatment with 0.89% H2SO4 (w/w) at 160 °C during 5 min, followed by an enzymatic hydrolysis, which released 75% of the sugars contained in corn stover. The hydrolysate was detoxified with the resin Dowex® Optipore® SD-2 and fermented by C. saccharobutylicum DSM 13864, producing 4.75 ± 0.25 g/L acetone, 9.02 ± 0.11 g/L butanol and 0.39 ± 0.01 g/L ethanol in 72 h, with a sugar consumption of 97.3 ± 0.27%. A two-stage gas stripping was applied to the fermentation broth, obtaining butanol-rich condensates (418–425 g/L in the organic phase) in a total time of 6 h.


Introduction
The development of a sustainable industry based on the use of renewable resources to produce fuels, chemicals and environmentally friendly materials has been recognised as a key issue of importance (Sarangi and Nanda, 2018). One of the alternatives that is gathering promising scientific efforts is the study of new fuels obtained from lignocellulosic biomasses, such as butanol production through acetone-butanolethanol (ABE) fermentative process (Nanda et al., 2017). In order to avoid the potential shortage of food that could occur by the switch from foodproducing to fuel-producing land, lignocellulosic wastes from agricultural industry have attracted much more attention as a sustainable alternative. Biobutanol production by ABE fermentation has already been successfully achieved from several agricultural wastes (Gottumukkala et al., 2017;Hijosa-Valsero et al., 2017;Hijosa-Valsero et al., 2018a,b) including corn stover (Qureshi et al., 2010). Currently, this lignocellulosic biomass is one of the most important feedstocks for bioethanol production and is considered a primary feedstock for biobutanol refinery in industrialized countries such as U.S. due to its immediate availability at the required industrial scale (Baral et al., 2017). In addition, corn stover also represents an interesting feedstock to biobutanol production due to its fibrous structure with a high carbohydrate content (about 18-22% hemicellulose and 32-36% cellulose). Both characteristics have attracted the attention of researches for using corn stover to produce second generation biobutanol (Malmierca et al., 2017). Aside from the selection of an appropriate lignocellulosic feedstock, biobutanol production requires the improvement of a number of technical procedures before fermentation, such as biomass pretreatment and hydrolysis of structural carbohydrates, selection of a suitable bacterial strain and optimization of the physicochemical conditions in the fermentation broth (pH, T, nutrients required, etc.) (Zhang et al., 2018). In addition, the pretreatment processes -because of sugars and lignin degradation-produce chemical by-products (phenolic compounds, furan derivatives and weak organic acids) which cause inhibition on solventogenic strains of Clostridium affecting biobutanol production (Grisales and Olivar, 2018). The production of these inhibitors is enhanced mainly under dilute acid pretreatments, particularly when using sulfuric acid (Baral and Shah, 2014). In order to deal with this drawback, several studies have proposed a wide range of alternatives to detoxify corn stover hydrolysates prior to bacterial fermentation. These methods include inhibitors extraction, adsorption, evaporation, electrodialysis, overliming, neutralization, steam stripping, as well as enzymatic and microbial treatments (Grisales and Olivar, 2018). Among the adsorption methods, the use of polymeric resins has been determined as an efficient method to detoxify biomassic hydrolysates, as well as a technique to remove alcohols from the fermentation broth (Carvalho et al., 2004;Gao and Rehman, 2016). In addition, polymeric resins potentiate the reduction of the overall production cost due to their reusability without affecting the process efficiency and in some cases without altering the sugar concentration contained in the hydrolysate (Gao and Rehmann, 2016). Previous studies on biobutanol production from corn stover have tried to solve each one of the abovementioned parameters through the selection of mutant and wild type solventogenic Clostridium strains Qureshi et al., 2010;Tang et al., 2017;Wang and Chen, 2011;Xue et al., 2016a;Zhang et al., 2018); the use of different pretreatment reagents and conditions (Baral and Shah, 2017;Tang et al., 2017;Xue et al., 2016a;Zhang et al., 2014) and hydrolysate detoxification by various methods Qureshi et al., 2010;Tang et al., 2017;Wang and Chen, 2011;Xue et al., 2016a;Zhang et al., 2018). Currently, studies conducted to potentiate biobutanol industrial production from corn stover are clearly required to increase yield values, but also to reduce the associated economic costs. Therefore, using alternatives such as an optimum biomass pretreatment with sulfuric acid or testing other detoxification techniques such as reusable resins, could decrease the economic and energetic requirements associated with the ABE fermentative process (Baral and Shah, 2017).
From an economic and energetic point of view, downstream processes such as solvent recovery and purification from fermentation broths are extremely important in biorefineries. The application of integrated recovery processes (which remove acetone, butanol and ethanol as they are being produced in the fermentation broth), like gas stripping, liquid-liquid extraction, vacuum fermentation, adsorption, reverse osmosis, pervaporation and perstraction, has been extensively reviewed (Abdehagh et al., 2014;Outram et al., 2016;Staggs and Nielsen, 2015;Xue et al., 2014;Xue et al., 2017). Recently, gas stripping has emerged as a very attractive alternative, since its energy requirements (including a final step of distillation) could be lower than previously calculated (Patraşcu et al., 2017;Xue et al., 2013a;Xue et al., 2013b;Xue et al., 2016b) and also due to configuration innovations like two-stage gas stripping (Xue et al., 2013a). Gas stripping has been usually tested as an integrated (in situ) recovery technique; but these types of operating conditions can increase the whole process energy requirements (Outram et al., 2016). It has been observed that gas stripping is also successful when applied once that the fermentation is finished, which requires much shorter stripping times (de Vrije et al., 2013). In view of the above, in this paper different strategies will be explored to deal with most of the major drawbacks that are present at each step of the ABE fermentative process when using corn stover as lignocellulosic feedstock. Consequently, acid pretreatment conditions will be refined in order to obtain suitable sugar and inhibitor concentrations in the hydrolysate, followed by a clostridial strain selection and the optimization of certain fermentation conditions (pH, T, CaCO3 and nutrient supplementation). The necessity of a detoxification process to remove inhibitors by inert adsorption resins prior to fermentation will be evaluated. Finally, the ABE solvents contained in the fermentation broth will be recovered by two-stage gas stripping. To the best of our knowledge, this is the first time that a complete ABE biorefinery process based on corn stover has been successfully developed, including a detoxification step with reusable resins.

Biomass description
Corn stover samples were obtained in November 2017 from experimental plots (ITACyL, Finca Zamadueñas, Valladolid, Spain). These agricultural by-products were dried in an oven at 45 °C until constant weight, ground in a rotary mill SM100 Comfort (Retsch GmbH, Haan, Germany) and sieved to a size of 0.5-1.0 mm. Moisture, ash, structural carbohydrates (cellulose and hemicellulose), Klason lignin, fats, proteins and total phenolic compounds were analysed as reported elsewhere (Hijosa-Valsero et al., 2017). Corn stover composition is shown in Table 1.  Figure A1). Physicochemical pretreatments were carried out with a 2-L high-pressure reactor made of alloy Carpenter-20 (Parr Instrument Company, Moline, IL, USA). Corn stover was immersed in an aqueous solution of sulfuric acid, with a solid-tosolvent ratio of 10% (w/w). Operation details of the reactor are described elsewhere (Hijosa-Valsero et al., 2018a). After the thermal pretreatment, an enzymatic hydrolysis was performed on the solid/liquid mixture obtained in the reactor according to the procedure described in section 2.4.

Pretreatment optimisation
The physicochemical pretreatment was optimised in order to obtain a fermentable broth with the maximum concentration of simple sugars (glucose, xylose, etc.) and the minimum concentration of inhibitors [formic acid, acetic acid, levulinic acid, furfural, 5-hydroxymethyl furfural (5-HMF) and phenolic compounds]. Although the analysis of sugars and inhibitors was performed after the combined and subsequent steps of physicochemical pretreatment and enzymatic hydrolysis, the process optimization was focused on physicochemical pretreatment conditions, because they are responsible for the generation of inhibitors and have a great influence on sugar release during enzymatic hydrolysis. In particular, the variables to be optimized were H2SO4 concentration (0.89, 1.13, 1.37, 1.60 and 1.84 % w/w) and temperature (125, 135, 145 and 160 °C). The treatment time in the reactor was experimentally set at 5 min, after checking that longer treatment times did not improve sugar release.

Enzymatic hydrolysis
After the thermal pretreatment, an enzymatic hydrolysis with Cellic CTec 2 was performed on the biomass solid/liquid mixture obtained in the reactor, following the method described by (Hijosa-Valsero et al., 2017). The employed dose of 36 µL/g biomass is equivalent to 3.78 FPU/g biomass. Spores from all the strains (except DSM 6228) were prepared and stored as explained in (Paniagua-García et al., 2018). For the asporogenic strain DSM 6228, lyophilised cells were resuspended in 10 mL of sterile Reinforced Clostridial Medium -RCM (Oxoid, Basingstoke, UK) supplemented with 10 g/L glucose, and incubated 24 h at 35 °C under anaerobic conditions. Then, 1.5 mL were mixed with 0.4 mL glycerol (80% v/v) in a cryogenic vial and stored at -80 °C until being used. Cellular reactivation and innocula preparation were performed in liquid RCM or in a potato based medium (in the case of DSM 2152 and DSM 792) as detailed by Hijosa-Valsero et al. (2018b). Bacterial cultures were incubated for 20-48 h at 35 ºC in order to obtain a density of 5·10 8 cells/mL as determined by counting in a Bürker chamber (Paul Marienfeld GmbH & Co. KG, Lauda-Königshofen, Germany).
2.6. Strain selection for the fermentation of corn stover hydrolysates Corn stover hydrolysates were obtained after a pretreatment with H2SO4 (125 °C, 5 min, 0.89% w/w acid) and a subsequent enzymatic hydrolysis. For fermentation tests, nondetoxified corn stover hydrolysates were filtered through filter paper (No. 1305, 73 g/m 2 , Filtros Anoia SA, Barcelona, Spain) and supplemented with the standard nutrients (S) listed in Table 2. Inoculation and medium preparation were performed as described by Paniagua-García et al. (2018). Fermentations were conducted at 35 °C, 100 rpm and 96 h in an Infors HT Minitron orbital shaker (Infors AG, Bottmingen, Switzerland). These experiments were performed in triplicate with the fifteen bacterial strains listed in section 2.5. Fermentation controls were prepared with aqueous solutions containing glucose and xylose mixtures at similar concentrations to those of corn stover hydrolysates, and supplemented with the abovementioned nutrients and salts. The most appropriate strain was selected for the next experiments. Table 2. Different nutrient supplementation and fermentation conditions applied to corn stover hydrolysates.

Optimisation of fermentation conditions
Once that the most efficient strain had been selected, the most adequate nutrients and fermentation conditions for that strain were determined. In the first place, a Plackett-Burman experimental design was proposed to select the necessary nutrients and their concentrations. Corn stover was pretreated with H2SO4 (125 °C, 5 min, 0.89% acid w/w) and a subsequent enzymatic hydrolysis. The Plackett-Burman design consisted of twelve experimental runs combining the maximum and minimum values established for the ten independent variables (ten nutrients: yeast extract, KH2PO4, K2HPO4, NH4Cl, MgSO4·7H2O, FeSO4·7H2O, MnSO4·H2O, ZnSO4·7H2O, cysteine and CaCO3). The hydrolysates were fermented under the conditions described in section 2.6 and butanol was measured as the response variable. The characteristics of this experimental design are shown in Table A1. The nutrients and their concentration ranges were selected according to literature data Qureshi and Blaschek, 1999;Qureshi et al., 2014;Ujor et al., 2016).
In the second place, after having established the essential nutrients for the selected strain, fermentation conditions (temperature, initial pH and CaCO3 concentration for pH control) were optimised via response surface methodology (RSM) with a Box-Behnken experimental design consisting of 3 factors, 1 replicate, 15 runs, 1 block and 3 central points (Table A4). Corn stover was pretreated with H2SO4 (125 °C, 5 min, 0.89% acid w/w) and a subsequent enzymatic hydrolysis, and then it was fermented using the nutrients and fermentation conditions listed in Table A4.

Detoxification
The objective of pretreatment and fermentation optimisation was to obtain high sugar concentrations and low inhibitor concentrations in order to directly ferment corn stover hydrolysates, thus avoiding an expensive detoxification step to remove inhibitors. However, the severe pretreatments applied to this lignocellulosic biomass could require a detoxification process. The polymeric adsorption resins Amberlite® XAD-4, Dowex® Optipore® L-493 and Dowex® Optipore® SD-2 were evaluated for the removal of inhibitors from corn stover hydrolysates. About 58 g of each polymeric resin were placed inside a 120-mL glass column (chromatography column with soldered porous plate No. 0, model 340634, 400 mm x 20 mm i.d.; Pobel, Madrid, Spain) and they were soaked with distilled water, reaching a total volume of 90 mL. The resins were conditioned following a modification of (Schwartz and Lawoko, 2010). In brief, the resin Amberlite® XAD-4 was washed three times with 90 mL methanol at 4 mL/min and then was left in methanol overnight. Afterwards it was washed with 900 mL distilled water at 10 mL/min. The resins Dowex® Optipore® L-493 and Dowex® Optipore® SD-2 were washed three times with 90 mL NaOH 1 M at 4 mL/min and then with 900 mL distilled water at 10 mL/min. For the detoxification step, corn stover (10% solid biomass) was pretreated with a 0.89% H2SO4 w/w aqueous solution during 5 min. Two distinct pretreatment temperatures (125 °C or 160 °C) had to be tested in order to generate different sugar concentrations in the hydrolysate (see section 3.4). Then, a subsequent enzymatic hydrolysis was applied. The hydrolysate was passed through filter paper (No. 1305, 73 g/m 2 , Filtros Anoia SA, Barcelona, Spain), and its pH was adjusted to 5.42 (optimal pH for fermentation). Then, the hydrolysate was fed into the column at a rate of 2 mL/min using a peristaltic pump (10 min residence time). The detoxified hydrolysate was collected and fermented with C. saccharobutylicum DSM 13864 under the optimal conditions (O) described in Table 2. The use of other strains for detoxified hydrolysates was experimentally discarded (see Appendix and Figure A2).

Gas stripping
In order to recover solvents (acetone, butanol and ethanol) from the fermentation broths, a two-stage gas stripping was applied according to Díez-Antolínez et al. (2018). Corn stover hydrolysate was obtained by treating 10% solid biomass in an aqueous solution of 0.89% H2SO4 (w/w) at 160°C during 5 min, followed by enzymatic hydrolysis and detoxification with the resin Dowex® Optipore® SD-2. The hydrolysate was fermented by C. saccharobutylicum DSM 13864 during 72 h using the optimised combination (O) described in Table 2. About 3.2 L of fermentation broth containing 5.02 g/L acetone, 9.02 g/L butanol, 0.42 g/L ethanol, 1.61 g/L acetic acid, 0.95 g/L butyric acid and 1.19 g/L total sugars, were subjected to gas stripping during 5 h under the following conditions: Tfeed = 60 °C, Trefrigeration = 5 °C and gas flow = 1.34 L/min. Five millilitres of Antifoam A (Fluka Analytical, Sigma-Aldrich, Steinheim, Germany) were added to the fermentation broth at the beginning of the stripping to avoid foam formation. The condensate was collected and it separated spontaneously in two phases (organic and aqueous). The aqueous phase was further subjected to another gas stripping process during 1 h and its condensate was collected. The performance of the gas stripping process was calculated based on solvent recovery (η) and selectivity ( (1) where αi is the selectivity for compound i (for instance, butanol; αB), xi is the mass ratio of metabolite i in the feed solution, yi is the mass ratio of metabolite i in the condensate, ηi is the percentage recovery efficiency for metabolite i (in this case, butanol; ηB), mic is the mass of metabolite i in the condensate (expressed in g) and miF is the mass of metabolite i in the feed solution. Gas stripping rates [g/(L·h)] were calculated as proposed by Truong and Blackburn (1984).

Statistical analyses
Samples were compared with a one-way ANOVA and Tukey's HSD test using the software Statistica 7 (StatSoft Inc., Tulsa, OK, USA). Plackett-Burman experimental designs and Box-Behnken RSM experimental designs were made with the software Minitab 16 (Minitab Inc., State College, PA, USA).

Composition of corn stover hydrolysates
The effects of temperature and H2SO4 concentration on corn stover pretreatment were assessed in order to increase sugar release and minimise the generation of fermentation inhibitors. Table 3 shows the chemical composition of various corn stover hydrolysates obtained with different pretreatment conditions. It was observed that temperature had a clear direct effect both on sugar release and inhibitor formation when acid concentration was kept constant at 0.89% w/w (Table 3). On the contrary, for a fixed temperature of 125°C, an increase in H2SO4 concentration did not improve sugar release, but slightly favoured the generation of inhibitors, specially formic acid, acetic acid, furfural and phenolic compounds (Table 3). These data suggest that a sulfuric acid concentration of 0.89% w/w (which is equivalent to ~0.54% v/v) could be sufficient for corn stover degradation. In fact, Gao and Rehman (2016) pretreated Phragmites australis with increasing concentrations of H2SO4 (0.5-2.0% v/v) and they observed that the use of higher acid concentrations did not release more sugars, but only more inhibitors. In addition, they reported that H2SO4 concentration influenced the type of phenolic compounds generated. The amount of total sugars released from corn stover ranged between 34.7 and 50.4 g/L, depending on the physicochemical pretreatment used (Table 3), which implies a sugar recovery efficiency of 48-75%. According to a review published by Baral and Shah (2014), the sugar yields obtained from corn stover employing different pretreatment methods vary between 42 and 66 g of fermentable sugars per 100 g of dry corn stover. In the present study, as corn stover contained 57.23 % carbohydrates (Table 1), that yield would reach ~43 g of fermentable sugars per 100 g of dry corn stover in the best case. In order to avoid inhibitory problems, it was decided to select the pretreatment conditions of 0.89% H2SO4 w/w, 125 °C and 5 min, because they offered acceptable sugar concentrations and low inhibitor concentrations. The total sugar concentration obtained (~40 g/L), moderate as it may seem, might guarantee successful ABE fermentations, as observed in previous works with cheese whey (Díez-Antolínez et al., 2016) or coffee silverskin (Hijosa-Valsero et al., 2018b), where initial sugar concentrations as low as 30-34 g/L enabled the generation of 7.0-8.5 g/L biobutanol.  (Tang et al., 2017). The selected strain in the present work (DSM 13864), belongs to the species C. saccharobutylicum, which has been reported to cope with mixed agricultural and waste-based substrates (Shaheen et al., 2000).

Optimisation of fermentation conditions for non-detoxified corn stover hydrolysates
The most adequate nutrients and fermentation conditions for C. saccharobutylicum DSM 13864 were evaluated as described in the Appendix (Tables A1-A7). The optimal working conditions, as well as the nutrient concentrations established in the Plackett-Burman and RSM experiments, have been summarised in Table 2 (optimised conditions -O). The calculated optimum answer established that it would be possible to produce 7.56 g/L butanol when working at 28 °C, pH 5.42 and 8.0 g/L CaCO3. In order to validate the model, this optimal point was experimentally tested by fermenting nondetoxified corn stover hydrolysates under these optimised conditions. The results of model validation yielded a butanol concentration of 5.38 ± 0.21 g/L, which is far from the estimated value (7.56 g/L). In addition, other fermentation experiments were performed on different days under similar circumstances obtaining variable butanol concentrations in the range of 4.49-6.38 g/L. This lack of repeatability might be related to the inhospitable nature of corn stover acid hydrolysate for bacteria (Qureshi et al. 2010), which could cause a selective pressure on microorganisms via toxic effects, that would lead to the survival of those organisms with a higher resistance to inhibitors, which may not be the best butanol producers. Even if the same spores batch of a bacterial strain is used in different experiments, not all the individuals are identical and, in addition, can suffer genetic changes during their development and reproduction causing a large amount of phenotypical variability inside this environment. This potential selective pressure generated by the toxic compounds could be faced by applying a detoxification step for this corn stover hydrolysate. In any case, under other circumstances (detoxified hydrolysates; Figure  A2) the optimised conditions (O) were proved to be superior to the standard conditions (S) for C. saccharobutylicum DSM 13864 and therefore they were maintained for further experiments. Table 4 shows fermentation performances of non-detoxified corn stover hydrolysates found in literature. The results from the present work are in agreement with other comparable studies employing batch fermentations, where butanol concentrations of 0-6.93 g/L were attained. The highest titers had been reported by Zhang et al. (2018), who employed a hydrolysate with a greater initial sugar concentration compared to the present work (52 vs. 40.5 g/L), but they did not report sugar consumption values or butanol yields. This satisfactory result could be due to the selection of the most appropriate strain for this specific hydrolysate (C. saccharobutylicum DSM 13864). In addition, the registered sugar consumption of 83% is notably high, and it is near the values observed in semicontinuous reactors with in situ butanol recovery (Parekh et al., 1988). This fact could be related to the relatively low initial sugar concentration in the broth and to the hypothetical formation of an extracellular polysaccharide by C. saccharobutylicum DSM 13864; although other non-polysaccharide forming strains, such as DSM 6423 and DSM 2152, had also shown sugar consumption values above 70% (Figure 1).  Table A8.

Detoxification of corn stover hydrolysates
Due to the presence of inhibitors, corn stover hydrolysate was not readily fermentable, and butanol production did not exceed the threshold of 4-6 g/L in spite of strain screening, nutrient optimisation and adjustment of fermentation conditions. Therefore, after physicochemical pretreatment (0.89% H2SO4 w/w, 125°C, 5 min) and enzymatic hydrolysis, corn stover hydrolysate was detoxified with three adsorption resins as described in section 2.8. Detoxification results can be found in Table 5. Most inhibitors were removed from the hydrolysates (especially 5-HMF and phenolic compounds), but sugars were also retained by the resins (14.7-17.8%), which implies an important loss of sugars for the subsequent fermentation process. These detoxified hydrolysates were fermented during 96 h with C. saccharobutylicum DSM 13864. According to Figure 2a, the detoxification process did not clearly enhance bacterial activity. In fact, butanol production in the best case (5.62 ± 0.02 g/L, with Amberlite® XAD-4) was similar to that of the non-detoxified hydrolysate (p > 0.05). In spite of the successful removal of inhibitors, ABE concentrations were not improved in a significant way, probably because of the decrease in sugar concentrations caused by the resin treatment (33.8-35.1 g/L initial sugars). As a consequence, it was checked whether a detoxified hydrolysate with higher sugar concentrations would produce greater ABE yields. For this experiment, a hydrolysate containing about 50 g/L total sugars, obtained with an aqueous solution of 0.89% H2SO4 (w/w) at 160°C during 5 min (Table 3), was chosen and detoxified with the three resins. Inhibitor concentrations were importantly reduced by the resins, while sugar concentrations decreased by 11.9-13.2% (Table 5). Therefore, fermentations were started with initial sugar concentrations of 43.7-44.4 g/L, using the optimised nutrients and fermentation conditions (O) described in Table 2. The fermentation was finished after 72 h. The non-detoxified hydrolysate was not fermentable, yielding barely 0.12 ± 0.04 g/L butanol ( Figure  2b). On the contrary, detoxified samples obtained butanol concentrations of 8.71 ± 0.03 g/L, 8.39 ± 0.23 g/L and 9.02 ± 0.11 g/L for resins Amberlite® XAD-4, Dowex® Optipore® L-493 and Dowex® Optipore® SD-2, respectively. In addition, butanol concentrations recorded for the sample detoxified with Dowex® Optipore® SD-2 were not significantly different (p > 0.05) from those of the control, which indicates that any potential inhibitory effects had been eliminated during detoxification. Therefore, the pretreatment at 160 °C aiming at releasing more sugars was successful for ABE fermentation, provided that a detoxification step is performed. This process implied a clear improvement in butanol production (from 5.38 ± 0.21 g/L to 9.02 ± 0.11 g/L butanol) and fermentation time (from 96 h to 72 h) in comparison to the previous values obtained before detoxification (section 3.3). The fermentation of the hydrolysate detoxified with resin Dowex® Optipore® SD-2 obtained 4.75 ± 0.25 g/L acetone, 9.02 ± 0.11 g/L butanol, 0.39 ± 0.01 g/L ethanol, 2.17 ± 0.07 g/L acetic acid and 1.26 ± 0.07 g/L butyric acid, with a sugar consumption of 97.3 ± 0.27%, a butanol yield YB/S of 0.222 ± 0.003 g/g and a butanol productivity WB of 0.125 ± 0.002 g/(L·h).
The three adsorption resins used in the present study had been previously assessed for the adsorption of phenolic compounds (Lin and Juang, 2009), for the in situ recovery of butanol from fermentation broths (Xue et al., 2016c) and for the detoxification of various lignocellulosic hydrolysates, but they had never been used for the detoxification of corn stover hydrolysates. Gao and Rehman (2016) used the resin Dowex Optipore L-493 to simultaneously adsorb inhibitors and butanol during the fermentation of an acid hydrolysate of Phragmites australis. The resin adsorbed 2.98% glucose, 4.85% xylose, 7.84% acetic acid, 33.3% 5-HMF, 77.8% furfural and 95.1% total phenolic compounds. Shukor et al. (2014) detoxified the acid hydrolysate of palm kernel cake with Amberlite XAD-4 and observed a reduction of 50% furfural and 77% 5-HMF, without losses of glucose or mannose. On the other hand, Ezeji et al. (2007) compared overliming and Amberlite XAD-4 as detoxification methods for corn fiber hydrolysates and concluded that overliming was superior, although the resin removed 60-80% furfural, 5-HMF and ferulic acid. Sugar losses were 2.99% for overliming and 10.72% for the resin. Weil et al. (2002) subjected a corn fiber hydrolysate to detoxification with resins Amberlite XAD-4 and XAD-7 and were able to remove 96% furfural. Our results are in agreement with the reported removal ranges in the case of furfural, 5-HMF and phenolic compounds, and are slightly higher in the case of acetic acid. However, sugar losses in the present study (12-18%) are clearly higher than those mentioned in literature (0-11%). Table 6 presents a summary of fermentation performances using detoxified corn stover hydrolysates found in literature. The most common detoxification techniques for this substrate are dilution with water, overliming, washing of the solid biomass after physicochemical pretreatment, alkaline peroxide and adsorption onto activated charcoal. To the best of our knowledge, this is the first time that polymeric resins are employed for the detoxification of corn stover hydrolysate. Butanol production in the present work (9.02 g/L) is similar or higher than that of other papers (0.36-14.5 g/L) and it is in the range of those experiments where fermentations were started with sugar concentrations below 45 g/L (7.1-11.5 g/L butanol). Steam explosion, dilute sulfuric acid, dilute NaOH, organosolv and deep eutectic solvents have been employed as physicochemical pretreatments for corn stover hydrolysis (Table 4 and Table 6). Dilute sulfuric acid pretreatment is frequent at an industrial scale (Baral and Shah, 2014). The concentration of H2SO4 used in the present work (0.89% w/w or ~0.54% v/v) is clearly below the values of 1-2% v/v reported in literature (Table 4 and Table 6), which constitutes an advantage from both an economic and environmental points of view. Moreover, the short pretreatment times (5 min) proposed would require less energy consumption. Current detoxification methods include overliming, evaporation, adsorption (onto resins or activated charcoal) or biological methods (use of peroxidases and laccases) (Baral and Shah, 2014). Adsorption resins have been regarded as an expensive detoxification method (Mussatto and Roberto, 2004). However, the possibility of immobilising resins inside columns to perform continuous processes and the recyclability of these materials by regeneration make them attractive for industrial applications. In addition, other detoxification methods employing alkali addition or activated carbon can reduce acetic acid, furfural and phenolic compounds concentration, but they can also induce salt formation, which can be detrimental for Clostridium species (Baral and Shah, 2014). Furthermore, detoxification techniques by washing the solid biomass before the enzymatic hydroysis imply the loss of the pentoses and other hemicellulosic sugars released during the physicochemical pretreatment. Moreover, in comparison to detoxification by washing (Table 4, Table 6), resins could entail water savings. Sainio et al. (2011) proved that the resin Amberlite XAD-16 (whose composition is similar to that of XAD-4) could be regenerated with a 50% ethanol aqueous solution in order to remove fermentation inhibitors such as furfural. Weil et al. (2002) proposed the use of adsorption resins as detoxification method in continuous processing systems and they confirmed the regenerability of the resin Amberlite XAD-4 with ethanol. Table 5. Concentrations of total sugars and inhibitors in corn stover hydrolysates after detoxification with three adsorption resins. Percentage changes are shown between brackets.  Therefore, if adsorption resins were used for inhibitor removal at industrial scale in ABE biorefineries, they could be regenerated by washing them with the same solvents recovered from the fermentation broth (acetone or ethanol), thus obtaining a solution rich in furans and phenolic compounds which could also have commercial value. 3.5. Two-stage gas stripping of fermentation broths The detoxified hydrolysate was fermented, and the fermentation broth was subjected to a twostage gas stripping process. During the first stage of the gas stripping, butanol concentration in the fermentation broth decreased from 9.02 g/L to 2.07 g/L in 5 h, and a condensate containing 169.26 g/L butanol was collected ( Table 7). This implies solvent recoveries (η) of 33.22% for acetone, 67.70% for butanol and 38.98% for ethanol, and selectivity factors (α) of 9.61 for acetone, 22.38 for butanol and 10.85 for ethanol. Gas stripping rates were 0.58 g/(L·h) for acetone, 1.39 g/(L·h) for butanol, 0.03 g/(L·h) for ethanol and 2.00 g/(L·h) for ABE. These results are in the upper range of previously reported values for gas stripping. Typical condensates of integrated fed-batch gas stripping processes contain 20-120 g/L acetone, 17-113 g/L butanol and 7-21 g/L ethanol (Cai et al., 2016;Xue et al., 2016b); with selectivity values of 0.45-30.5 for butanol and 4-30.5 for ABE; and gas stripping ABE rates of 0.02-1.34 g/(L·h) (Qureshi and Blaschek, 2001;Abdehagh et al., 2014;Xue et al., 2016b). The gas stripping process was successful in spite of the presence of an antifoaming agent, which is known to reduce the system performance (Ezeji et al., 2005). The relatively high butanol concentration in this condensate, notably above butanol solubility in water (7.4 g/100g at 298 K; Barton, 1984), resulted in the spontaneous separation of an aqueous and an organic phase. The organic phase had a butanol concentration of 425.35 g/L. The aqueous phase, with a butanol concentration of 88.70 g/L, was collected and subjected to a second gas stripping stage.
The second stage of the gas stripping caused a decrease in butanol concentration in the feed solution from 88.70 g/L to 1.60 g/L in 1 h. In this case, a condensate containing 263.29 g/L butanol was collected (Table 8). The recorded solvent recoveries (η) were 45.81% for acetone, 73.38% for butanol and 70.01% for ethanol. In this second gas stripping step, selectivity factors (α) were lower than in the first stage, since solvent concentrations in the initial feed solution were much higher at the beginning of the second stripping. Thus, selectivity values (α) of 1.94 for acetone, 3.67 for butanol and 2.86 for ethanol were attained. Gas stripping rates during this second stripping stage were 46.9 g/(L·h) for acetone, 87.1 g/(L·h) for butanol, 3.79 g/(L·h) for ethanol and 27.6 g/(L·h) for ABE. It has been reported that condensates of two-stage gas stripping processes contain 119-198 g/L acetone, 337-451 g/L butanol and 21-23 g/L ethanol (Cai et al., 2016). Once more, the condensate spontaneously separated in two phases. The aqueous phase contained 129.67 g/L butanol, a value which is above its solubility in water, but this fact could be explained by the presence of 99.54 g/L acetone in this aqueous phase, which could have favoured butanol solubility (Xue et al., 2013a). On the other hand, the organic phase contained 417.71 g/L butanol, which is a high value similar to that obtained in the organic phase of the first stripping stage. Figure 3 summarises butanol fate during the different steps of the gas stripping process. Total mass recoveries (sum of organic phase 1, organic phase 2 and aqueous phase 2) reached 19.91% acetone, 60.28% butanol and 30.37% ethanol.

Conclusions
It is possible to pretreat corn stover by employing low H2SO4 concentrations (0.89% w/w) and yet release about 75% of its constitutive carbohydrates after enzymatic hydrolysis. However, high temperatures (160 °C) are necessary to guarantee an efficient pretreatment. Due to the presence of fermentation inhibitors in the hydrolysate, a correct strain choice and a detoxification step are essential to carry out a successful ABE fermentation. Despite the fact that in this study one of the highest butanol values has been reported without the necessity of using genetically improved bacterial strains, it could be interesting to test the whole butanol production workflow proposed in this paper with some hyerproducing solventogenic strain previously described, for instance C. beijerinckii P260. The use of polymeric adsorption resins immobilised in columns for the detoxification is an attractive alternative, since they may be regenerated and reused by washing them with organic solvents (ethanol, acetone); which could imply the simultaneous recovery of retained phenolic compounds and furans. The present method (including pretreatment, detoxification and fermentation) enabled the production of 43.9 g acetone/kg corn stover, 83.4 g butanol/kg corn stover and 3.61 g ethanol/kg corn stover. These solvents could be recovered from the fermentation broth by two-stage gas stripping, producing condensates with 46-91 g/L acetone, 169-263 g/L butanol and 4.5-13.2 g/L ethanol, which separated spontaneously into an aqueous and an organic phase, the latter containing 38-99 g/L acetone, 418-425 g/L butanol and 4.2-13.9 g/L ethanol.

Plackett-Burman experimental design
The most adequate nutrients and fermentation conditions for C. saccharobutylicum DSM 13864 were evaluated. The results of the Plackett-Burman experiments for nutrient optimisation are shown in Tables A1, A2 and A3. Butanol concentrations for the twelve Plackett-Burman experiments ranged between 0.11 and 6.21 g/L, depending on the nutrient composition (Table A1). The best fitting was obtained when MnSO4·7H2O, FeSO4·7H2O, ZnSO4·7H2O and CaCO3 were removed from the model (Table A2, Table A3). There were three variables with a negative effect (negative sign, Table A2) which were cysteine, ZnSO4·7H2O and CaCO3. These variables should be set at the lowest value of their respective range for further experiments. For cysteine and ZnSO4·7H2O, their lowest value is zero (Table A1) and as a consequence they should be removed from the nutrient cocktail. Regarding CaCO3, its lowest value in the experiment was 1 g/L (Table A1). It would be interesting to figure out which would be the most appropriate concentration of CaCO3 for the fermentation; therefore, the optimal concentration of this salt will be assessed in the RSM experiment. The nutrients yeast extract, KH2PO4, K2HPO4 and NH4Cl had a positive and significant effect (Table A3); thus these compounds were selected and their concentrations were established at their highest range value for the next experiments. The salts MnSO4·7H2O, MgSO4·7H2O and FeSO4·7H2O had a positive (but not significant, p > 0.05) effect, and it was decided to keep them in the nutrient cocktail at the highest value in their concentration range. According to these reasons, yeast extract, KH2PO4, K2HPO4, NH4Cl, MgSO4·7H2O, MnSO4·7H2O and FeSO4·7H2O were included in the fixed nutrient cocktail which would be used in the RSM experiment (see concentrations in Table A4).

Response surface methodology (RSM)
In the RSM experiment, fermentation temperature, initial pH and CaCO3 concentration were optimised (Table A4). Butanol concentrations under the fifteen tested conditions from the RSM ranged between 0 and 6.55 g/L (Table A5). According to regression coefficients (Table A6) and analysis of variance (Table  A7), the model explained 87.75% of the variation, but its predicted R-square value of 0% indicated that the model was overfit. The removal of non-significant variables (with p > 0.05) to generate new models did not improve R values (data not shown). Therefore, the information from Table A6 and Table A7 was used to create an equation to estimate butanol concentrations from independent variables: where T is the fermentation temperature (°C), pH is the initial pH value and CaCO3 is the concentration of calcium carbonate (g/L) added to the fermentation broth. This equation was used to mathematically determine the optimal values of temperature, pH and CaCO3 concentration which would maximise butanol production. The calculated optimum answer established that it would be possible to produce 7.56 g/L butanol when working at 28 °C, pH 5.42 and 8.0 g/L CaCO3. These optimal conditions, as well as the nutrient concentrations established in the Plackett-Burman experiment, have been summarised in Table 2 (optimised conditions -O). In order to validate the RSM model, this optimal point was experimentally tested by fermenting corn stover hydrolysates under these optimised conditions ( Table 2). The results of model validation yielded a butanol concentration of 5.38 ± 0.21 g/L, which is far from the estimated value (7.56 g/L). In addition, other fermentation experiments were performed on different days under similar circumstances obtaining variable butanol concentrations in the range of 4.49-6.38 g/L.

Detoxification and choice of strains for detoxified hydrolysates
As the nature of corn stover detoxified hydrolysates is different from that of non-detoxified hydrolysates, it was decided to check whether the selected strain, additional nutrients and fermentation conditions established for the non-detoxified hydrolysates were also appropriate for the detoxified hydrolysate. Therefore, the three best performing strains for the non-detoxified hydrolysate (DSM 13864, DSM 6423 and DSM 2152; see section 3.2) were compared to ferment the detoxified hydrolysate of the resin Dowex® Optipore® SD-2, under the two different conditions shown in Table 2 (S and O). This hydrolysate had been prepared with 0.89% H2SO4 w/w, 5 min, 160 °C. The results of this strain comparison are given in Figure A2.
The highest butanol concentration was obtained by DSM 13864 with optimised conditions (8.55 ± 0.07 g/L), followed by DSM 13864 with standard conditions (7.64 ± 0.03 g/L), DSM 2152 with standard conditions (7.17 ± 0.11 g/L) and DSM 6423 with standard conditions (6.44 ± 0.21 g/L). The nutrient composition and fermentation conditions that had been optimised for DSM 13864 in the non-detoxified hydrolysate (O) were the most efficient (p < 0.05) for DSM 13864 with the detoxified hydrolysate too. On the contrary, standard conditions (S) were the most appropriate for strains DSM 6423 and DSM 2152 with the detoxified hydrolysate (p < 0.05). In summary, with this detoxified hydrolysate, butanol production with DSM 13864 and Optimised conditions is statistically superior (p < 0.05) to that obtained by DSM 6423 and DSM 2152 under conditions O or S. Therefore, DSM 13864 and its optimised nutrient and fermentation conditions are also valid for detoxified corn stover hydrolysate.  Table 2). Table A8. Fermentation of corn stover hydrolysates before and after detoxification with three adsorption resins with C. saccharobutylicum DSM 13864, under two different pretreatment conditions. Note: See Figure 2 for more details.