Journal of Membrane Science 487 (2015) 40–50 Contents lists available at ScienceDirect Journal of Membrane Science journal homepage: www.elsevier.com/locate/memsci Assessing nanofiltration and reverse osmosis for the detoxification of lignocellulosic hydrolysates N. Nguyen a,b,c,1, C. Fargues a,b,c, W. Guiga a,b,c, M.-L. Lameloise a,b,c,n a AgroParisTech, UMR Ingénierie Procédés Aliments, 1 avenue des Olympiades, F-91300 Massy, France INRA, UMR Ingénierie Procédés Aliments, F-91300 Massy, France c Cnam, UMR Ingénierie Procédés Aliments, F-75141 Paris, France b art ic l e i nf o a b s t r a c t Article history: Received 23 December 2014 Received in revised form 24 March 2015 Accepted 28 March 2015 Available online 6 April 2015 During hydrolysis of lignocellulosic materials for ethanol production, compounds toxic for fermentation are formed. Ten nanofiltration (NF) and reverse osmosis (RO) membranes with low molecular weight cut-off (150–400 g mol 1) were screened on a flat-sheet plant for their ability to separate C5 and C6 sugars from acetic acid, furfural, 5-hydroxymethyl furfural and vanillin in a model solution. RO led to the highest sugars rejection ( 497%) but inhibitors transmission was low. NF membranes, especially NF270, NF- and NF245 (Dow) and DK (GE Osmonics) were found suitable for detoxification with glucose rejection 494% and inhibitors transmission 480%. At high Volume Reduction Ratio, VRR, transmission of inhibitors was still enhanced ( 496% at VRR ¼8 and 10 bars). In these conditions, NF270 gave the highest permeate flux (20 L h 1 m 2) followed by DK, NF- and NF245. However, DK and NF- could be preferred because of lower sugar loss. & 2015 Elsevier B.V. All rights reserved. Keywords: Nanofiltration Reverse osmosis Detoxification Inhibitor Lignocellulosic hydrolysate 1. Introduction Lignocellulosic biomass is currently being considered as a new renewable source of energy for the production of second generation bioethanol. Carbohydrate content can be converted into fermentable sugars directly by acid hydrolysis or indirectly by a two-stage process involving pretreatment and enzymatic hydrolysis. Although diluteacid hydrolysis is a fast and cheap method for obtaining sugars from lignocellulosic materials, it leads to the formation of toxic compounds for fermentation such as furan derivatives (furfural and 5hydroxymethyl furfural (HMF)), aliphatic acids (mainly acetic, formic and levulinic acids) and phenolic compounds. Such inhibitory substances adversely affect the productivity and the yield of ethanol fermentation [1]. In order to enhance the effectiveness of fermentation, sugars concentration should be increased and inhibitors should be removed. Various detoxification methods have been reviewed ([2–4]). The most extensively studied are based on physical and chemical principles, such as evaporation, overliming, solvent extraction, adsorption and ion-exchange. Biological methods also recently n Corresponding author at: AgroParisTech, UMR Ingénierie Procédés Aliments, 1 avenue des Olympiades, F-91300 Massy, France. Tel.: þ 33 1 69 93 50 76. E-mail addresses: nhunguyen@gmail.com (N. Nguyen), marie-laure.lameloise@agroparistech.fr (M.-L. Lameloise). 1 Present address: Lac Hong University, 10 Huynh Van Nghe, Bien Hoa, Dongnai, Vietnam. http://dx.doi.org/10.1016/j.memsci.2015.03.072 0376-7388/& 2015 Elsevier B.V. All rights reserved. appeared based on the bioconversion of inhibitors into less toxic compounds. So far, however, none of these treatments proved its ability to remove all families of inhibitors and each of them has its own drawbacks: high processing costs (evaporation), high chemicals consumption and production of wastes (overliming, ion-exchange, adsorption), hazardous solvent handling (liquid/liquid extraction), significant sugar loss or degradation (overliming), and low efficiency (biological methods). Moreover, with the exception of evaporation, they do not allow simultaneous detoxification and concentration of sugars. Pressure-driven membrane technology has already shown advantages in various fields of biorefinery as compared to other separation and purification techniques, including lower energy consumption, sustainable processing and flexibility. However, regarding the detoxification of lignocellulosic hydrolysates for the production of second-generation ethanol or other bioconversions, pressure-driven membranes have been considered only recently and the first review addressing their potential in this particular field is that of Abels et al. [5]. Actually, major inhibitors have lower molecular weight (MW) than sugars (formic acid: 46; acetic acid: 60; levulinic acid: 116; furfural: 96, HMF: 126 g mol 1 compared to 150 and 180 for C5 and C6 sugars, respectively). Thanks to size exclusion effects, membranes with small molecular weight cut-off (MWCO of around 150 g mol 1) can be expected to let inhibitors as acids, furfural and HMF pass through while retaining sugars in the retentate. Moreover, at the low pH of the N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 hydrolysates (pH E3), acids are mostly in their undissociated form (only 1.7% of acetic acid is dissociated) and membrane charge density is low; electrostatic repulsion is therefore minimized. Transmission of phenolic inhibitors is more questionable because they have higher MW than the smallest monosaccharide (for example vanillin: 152, vanillic acid: 168, syringaldehyde: 182, ferulic acid: 194, syringylpropane: 196 g mol 1). However, most of them have a marked hydrophobic character as shown by octanol–water partition coefficients KOW higher than 41.5 [6]. Hydrophobic compounds may show lower rejection than could be predicted from size-exclusion mechanisms ([7,8]); this would be related to enhanced adsorption on the surface which facilitates transport through the membrane ([9,10]). The potential of nanofiltration (NF) for this particular application was first demonstrated by Weng et al. [11] with GE Osmonics Desal 5 DK membrane on a xylose–acetic acid mixture and further confirmed on rice straw hydrolyzate [12]. The presence of sugars seems to decrease acetic acid rejection to even negative values. With Desal 5 DK and Alfa-Laval-NF membranes, Zhou et al. [13] observed rejections from 85–89% for xylose to 96–98% for glucose and confirmed negative values for acetic acid rejection. Encouraging observations were reported by Qi et al. [14] for furfural removal by Dow NF90 and NF270 from a glucose–xylose mixture but the experimental set-up (a dead-end filtration cell with 4.5 cm2 filtration area and magnetic stirring) was far from crossflow conditions. They are few published results relating to phenolic compounds. Maiti et al. [15] tested flat polyamide membranes with MWCO between 100 and 400 g mol 1 and a spiral-wound PES membrane with a 150 g mol 1 MWCO on a synthetic mixture of mono- and di-saccharides and several inhi- 41 bitors including vanillic and ferulic acids: high transmission of phenolics was observed. With reverse osmosis (RO), quite complete sugar recovery may be expected but perhaps at the expense of detoxification efficiency. Not much work can be found. One of them is with a model solution of acetic acid, xylose and glucose and Alfa-Laval RO98pHt and RO99 membranes [13]. Rejection close to 100% was found for sugars at 30 bars but detoxification was limited with rejection of about 45% for acetic acid. This is consistent with the results of Sagne et al. ([16,17]) on detoxification of beet distillery condensates containing similar inhibitory compounds: acetic acid and furfural rejections were found less than 50% with Hydranautics CPA2 membrane at similar pressure. Higher transmissions could be achieved at lower pressure, with the drawback of lower permeate flux. The aim of this work was to screen a large panel of NF and RO membranes on a flat-sheet laboratory plant for their ability to separate inhibitors from sugars. This was done on a complex model solution simulating the average composition of a dilute acid hydrolysate containing three sugars: glucose, xylose and arabinose and inhibitors of various chemical families: acetic acid as major inhibitor of the carboxylic acids family, furfural and HMF as furan derivatives and vanillin as phenolic compound. For each membrane, effect of transmembrane pressure and concentration on permeate flux and solute rejection was studied. Membranes and operational conditions providing the highest sugar rejection together with the highest inhibitor transmission were selected for future pilot-scale studies and fermentation evaluation. 2. Materials and methods Table 1 Characteristics of the solutes used in model hydrolysate (sugars are represented under their dominant form in water). Structure MW (g mol pKa Stokes diameter log KOW (nm) 1 ) Glucose 180 12.28 [12] 0.726 [12] 3.24 [19] Xylose 150 12.15 [12] 0.638 [12] 1.98 [8] 2.1. Model solution Model solution was chosen based on a literature survey of hydrolysates compositions ([18]). It contained xylose (15 g L 1), glucose (10 g L 1), arabinose (5 g L 1), acetic acid (5 g L 1), 5hydroxymethylfurfural (1 g L 1), furfural (0.5 g L 1) and vanillin (0.05 g L 1). Chemicals were purchased from Sigma-Alldrich (St Quentin Fallavier, France) and Interchim (Montluçon, France). pH of model solution was 3, close to the pH of real hydrolysates. Solutes characteristics are given in Table 1. 2.2. Membrane selection Arabinose 150 – 0.635 [12] Acetic acid 60 4.75 0.412 [12] 0.17 [6] HMF 126 4 12 [15] 0.463 [12] 0.37 [6] Furfural 96 4 12 [15] 0.438 [12] 0.41 [6] Vanillin 152 8.2 [20] – – 1.21 [6] Ten commercially available RO and NF membranes (Table 2) were selected from literature results including own research on condensates detoxification [25] and data from suppliers. For NF membranes, MWCO was in the range of 150–400 g mol 1 as given by manufacturers. MWCO is indicative because determination methods may vary from one manufacturer to the other. All membranes were thin-film composite membranes with a polyamide active layer. Fully aromatic polyamide is used for NF90 and RO membranes and mixed aromatic/aliphatic polyamide (polypiperazine amide) for NF. Membranes may also undergo specific and proprietary treatments, such as blending with unreactive polymers to change hydrophobicity and density of the top layer or surface grafting. Maximal operating conditions are 45–50 1C and 41 bar (except NF245: 54.8 bar) and 2–3 to 10–11 for pH. Virgin membranes were first dipped in KOH solution (0.4 g L 1) to remove storage chemicals and then flushed in desionized water for at least 24 h until tested in the pilot. After each filtration experiment, the membranes were cleaned with KOH (0.4 g l 1) under low pressure and high flow rate and rinsed many times with desionized water in order to recover the initial 42 N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 Table 2 Characteristics of studied RO and NF membranes. RO NF Membrane Supplier Active layer material Rejection (from supplier) CPA2 CPA3 ESPA2 XLE SG NF90 Hydranautics Fully aromatic polyamide 4 99.5% NaCl 4 99.6% NaCl 4 99.5% NaCl 99% NaCl 4 97% NaCl 4 97% MgSO4 Dow Filmtec GE Osmonics Dow Filmtec NF270 NFNF245 DK Polypiperazine amide GE Osmonics 4 97% MgSO4 4 98% MgSO4 – 4 98% MgSO4 hydraulic permeability. 0.1 M sodium bisulfite solution was used to prevent bacterial growth during long-term storage. 2.3. Filtration device and protocol Experiments were run on a LabStak M20 filtration device (AlfaLaval, Les Clayes sous Bois, France) allowing several flat-sheet membranes to be tested in series simultaneously. Permeate could be collected separately for each membrane but the current retentate only was available. Filtration area was 0.036 m2 (2  0.018 m2) for each membrane type. All experiments were run at 20 1C and 400 L h 1 feed flow rate. Pure water was first filtered for membrane compaction at 20 bar transmembrane pressure (TMP) until flux stabilization. Pure water flux JW and permeate flux JP through the membrane are calculated by permeate flow-rate measurements as J w ; J P ¼ F P =S ð1Þ 1 where Fp is the permeate flow rate (L h ) and S is the membrane area (m2). Solution–diffusion model as described in 1965 by Lonsdale et al. [26] is commonly applied for RO membranes. We assume that it can also be extended to tight NF membranes (as studied here). In this model, JW is supposed proportional to the transmembrane pressure (TMP) applied, according to
 1 J W ¼ AW UTMP ð2Þ Lh m 2 where AW is the permeability of the membrane to water (L h 1 m 2 bar 1). Water permeability was then calculated at 400 L h 1 feed flow rate by increasing TMP from 5 to 30 bars by 5 bars and calculating the corresponding water flux Jw through permeate flow rate measurements. The effect of pressure on rejections and permeate flux was studied on NF and RO groups separately in batch recycling mode (retentate and permeates recycled in the feed tank). Pressure was increased from 5 to 30 bars by 5 bars. At each pressure, after 30 min circulating for stabilization, samples of feed, current retentate and individual permeates were collected and permeate flux was calculated. Effect of concentration was studied by operating the system in the concentration mode, where retentate was recycled in the feed tank, while permeate was extracted until the desired Volume Reduction Ratio (VRR), defined as VRR ¼ VF VF P VP ð3Þ Pore diameter (nm) MWCO (from supplier) (g mol – – 0.73 [21] 0.68 [22] 0.84 [22] 0.78 [23] – 1.25 [21] 0.96 [24] 200–400 1 ) 200–400 200–400 o 300 150–300 P where VF is the initial feed volume and V P the total permeate volume extracted till then. A feed volume of about 17 L was used. Pressure was 30 bars for RO membranes and 10 bars for NF membranes. At each VRR, the system was let to stabilize for 30 min before sampling feed, current retentate and each permeate. Permeate flux was calculated as above (Eq. (1)). Observed rejection Ri and transmission Ti were calculated whatever the filtration mode as   C P;i ð%Þ ð4Þ Ri ¼ 100 U 1 C F;i T i ¼ 100 Ri ð5Þ where CP,i and CF,i are the concentrations of solute i (g L 1) in the permeate and in the feed tank, respectively. In concentration studies (VRR increase), CF,i increases and is measured accordingly for rejection calculation. 2.4. Analytical methods Samples collected during the experiments were analyzed by high performance liquid chromatography (HPLC). The system was composed of a 321 pump (Gilson, Roissy, France), a Degasys DG-1310 degassing system (Uniflow, Tokyo, Japan), a Biotek Kontron Instruments 465 automatic autosampler (Gilson, Roissy, France) and an Igloo-cil oven to control column temperature (Cluzeau Info Labo, Courbevoie, France) (70.8 1C). Data were acquired and processed by Empower software (Waters, Guyancourt, France). Sugars concentration (glucose, xylose, arabinose) was quantified on a Nucleodur 100-5 NH2-RP column heated at 30 1C and equipped with a refractometric detector (Waters 410, Waters, Guyancourt, France). Mobile phase was acetonitrile:water (85:15) at a flow-rate of 0.7 mL min 1. Inhibitors (acetic acid, HMF, furfural and vanillin) were quantified on a Betamax Neutral column (150 mm  4.6 mm i.d., 5 mm particle size; Thermo-Electron Corporation, Courtaboeuf, France) heated at 50 1C and a Waters 996 photodiode array detector (Guyancourt, France) operating at 207 nm wavelength for acetic acid and 249 nm for HMF, furfural and vanillin. The mobile phases for the elution gradient were (A) H2SO4 5.10 4 mol L 1 and (B) acetonitrile and flow rate was 1 mL min 1 as already optimized for similar solutes [17]. The gradient consisted of an increase of B from 5% to 40% in 10 min. It then returned in 1 min to 5% after a 5 min plateau. After each run, the column was equilibrated under the starting conditions for 10 min. N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 43 350 XLE NF270 ESPA2 200 NF90 300 CPA2 DK 250 SG CPA3 JW (L h-1 m-2) JW (L h-1 m-2) 150 100 NFNF245 200 150 100 50 50 0 0 0 5 10 15 20 25 30 TMP (bar) 0 5 10 15 Table 3 Pure water permeability of studied RO and NF membranes. Membrane AW (L h RO CPA3 SG CPA2 ESPA2 XLE 2.6 2.7 3.1 5.8 7.7 NF NF245 NFDK NF90 NF270 3.7 5.6 6.1 6.2 12.0 25 30 TMP (bar) Fig. 1. Pure water flux for studied RO (a) and NF (b) membranes (20 1C; feed flow rate¼ 400 L h Group 20 1 m 2 bar 1 ) 3. Results and discussion 3.1. Water permeability Permeability values are calculated from the slopes of the curves JW ¼f(TMP) (Fig. 1) and Eq. (2). Results are given in Table 3. In the group of RO membranes, XLE exhibits the highest permeability followed by ESPA2. Actually, ESPA2 is known to be a loose membrane with relatively high permeability related to its corrugated surface and resulting in a doubling of the water flux compared to CPA2 membrane [27]. Its permeability, about 6 L h 1 m 2 bar 1, is similar to that for NF90 and DK nanofiltration membranes, for which values are consistent with results published by other authors ([22,28]). The lowest permeability values are found for CPA2, CPA3 and SG membranes, similar to that of NF245. The highest permeability of NF270 among NF membranes might be related to its low active layer thickness compared to DK, for example (90 nm for NF270 against 120 nm for DK according to Dalwani et al. [28]). Actually, RO membranes do not prove systematically less permeable to water than NF. Rejections for both membrane categories might be surprising and difficult to predict. 3.2. Effect of pressure on rejections Rejections are plotted versus permeate flux rather than TMP to allow direct comparison between membranes of different permeabilities (Figs. 2 and 3). The six successive points for a given membrane thus correspond to TMP¼ 5–30 bars by 5 bars steps. 1 ). 3.2.1. RO membranes Due to its very high rejection performances ( 497% for sugars and 480% for HMF and vanillin), NF90 is presented here with the RO membranes group (Fig. 2). For those membranes, except at 5 bars for XLE and CPA type membranes, rejection of sugars is always higher than 95% probably due to a predominant size exclusion effect. As expected, inhibitors rejections increase with pressure or permeate flux according to the solution–diffusion model and a plateau is achieved at different pressures depending on the membrane. Vanillin rejection is high, always above 60% and the plateau value achieved between 10 and 20 L h 1 m 2 is above 86% with low differences between the membranes. Steric exclusion is probably the dominant mechanism for its rejection by RO membranes. Regarding the other inhibitors (HMF and especially acetic acid and furfural), significant difference is observed between the membranes. Compared at an average permeate flux for RO membranes of about 18 L h 1 m 2, XLE shows the highest rejections of acetic acid, furfural and HMF (80% for acetic acid, 85% for furfural and 98% for HMF). Intermediate rejections are observed with ESPA2, SG and NF90, namely 50% for furfural, 60% for acetic acid and 80% for HMF. Finally, the lowest rejections (or highest transmissions) are obtained for the CPA group with 25% for furfural, 40% for acetic acid and 60% for HMF. These observations are consistent with literature data. According to Bennani et al. [21], XLE has the majority of pore diameters in the range 0.55–0.85 nm with a symmetrical distribution around 0.7 nm, whereas CPA3 has pore diameter around 0.9 nm, approaching the nanofiltration-type membrane. Better rejections with ESPA2 as compared to CPA had already been observed by Fargues et al. [29] and related to the higher cross-linking of the aromatic polyamide layer in ESPA2 (as deduced from zeta potential measurements) corresponding to higher polymer density and hindered diffusion of the solutes. For all membranes except XLE, the orders of rejections and of molecular weights do not match exactly: furfural with MW ¼96 g mol 1 is less rejected than acetic acid (MW¼60 g mol 1). It can therefore be speculated that affinity with the membrane material also plays a role in transmission. Furfural with log KOW ¼ 0.4 has probably a higher affinity with the aromatic polyamide active layer of the membrane through π–π interaction ([29,30]) and goes through the membrane much more easily than the highly polar acetic acid (log KOW ¼ 0.17) which does not interact with the polyamide material. Vanillin has an even larger log KOW (1.21) but no permeation is possible as it is N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 100 100 80 95 60 R glucose (%) R furfural (%) 44 XLE ESPA2 SG NF90 CPA2 CPA3 40 20 XLE ESPA2 SG NF90 CPA2 CPA3 90 85 80 75 0 0 10 20 30 40 50 60 0 10 20 Jp (L h-1m-2) 40 50 60 100 80 60 R xylose (%) R acetic acid (%) 100 XLE ESPA2 SG NF90 CPA2 CPA3 40 20 95 XLE ESPA2 SG NF90 CPA2 CPA3 90 85 80 75 0 0 10 20 30 40 -1 50 60 0 10 20 -2 Jp (L h m ) 30 -1 40 50 60 -2 Jp (L h m ) 100 R arabinose (%) 100 80 R HMF (%) 30 Jp (L h-1m-2) XLE ESPA2 SG NF90 CPA2 CPA3 60 40 20 95 XLE ESPA2 SG NF90 CPA2 CPA3 90 85 80 75 0 0 10 20 30 40 -1 50 60 -2 0 10 20 30 -1 Jp (L h m ) 40 50 60 -2 Jp (L h m ) R vanillin (%) 100 80 XLE ESPA2 SG NF90 CPA2 CPA3 60 40 20 0 0 10 20 30 -1 40 50 60 -2 Jp (L h m ) Fig. 2. Sugars and inhibitors rejection versus permeate flux in the recycling mode with RO and NF90 membranes (TMP increased from 5 to 30 bars by 5 bars step; 20 1C; feed flow rate¼ 400 L h 1). mostly excluded by steric effect. For XLE membrane, rejection follows the order of molecular weight of the inhibitors showing that with this tight membrane, size exclusion is probably the dominant effect for the MW range investigated (60 g mol 1 oMW o152 g mol 1). For detoxification purposes, the lowest rejection of inhibitors is required. It is obtained for CPA membranes but, even at the minimal pressure necessary for the highest sugars rejection (10 bars) it is still above 40% and 75% for HMF and vanillin, respectively. Moreover, when increasing pressure to improve permeate fluxes, inhibitors rejection will increase. 3.2.2. NF membranes At TMP above 10 bars, a high rejection ( 494%) is observed for glucose whatever the NF membrane (Fig. 3). For xylose, the plateau is achieved at 15 bars with 90% rejection for all membranes except NF270, which shows lower rejection (83%). Although sugars rejections are high, they are a little lower than with RO and a difference between C6 and C5 sugars is now visible. At equivalent permeate flux, such difference between glucose and xylose rejection was also reported by Sjöman et al. [31] with DK and NF270. At TMP ¼10 bars, permeate flux was 76 L h 1 m 2 for NF 270, around 43 L h 1 m 2 for NF and DK and 31 L h 1 m 2 for N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 NFNF245 DK NF270 R furfural (%) 20 15 10 100 R glucose (%) 25 90 80 NFNF245 DK NF270 70 5 60 0 -5 0 50 10 0 15 0 -1 0 200 50 -2 NFNF245 DK NF270 20 15 10 150 -1 5 200 -2 m ) NFNF245 DK NF270 100 R xylose (%) R acetic acid (%) 25 90 80 70 60 0 0 50 100 15 0 -1 0 200 50 -2 J p (L h m ) 100 Jp (L h NFNF245 DK NF270 20 15 10 5 -1 150 200 -2 m ) 100 R arabinose (%) 25 R HMF (%) 10 0 Jp (L h Jp (L h m ) 90 80 NFNF245 DK NF270 70 60 0 0 50 10 0 Jp (L h 15 0 -1 200 -2 m ) 25 R vanillin (%) 45 0 50 100 Jp (L h -1 150 200 -2 m ) NFNF245 DK NF270 20 15 10 5 0 0 50 100 Jp (L h -1 15 0 200 -2 m ) Fig. 3. Sugars and inhibitors rejection versus permeate flux in the recycling mode with NF membranes (TMP increased from 5 to 30 bars by 5 bars step; 20 1C; feed flow rate ¼400 L h 1). NF 245. At very high permeate flux as reached with NF 270 at TMP ¼20 bars (147 L h 1 m 2 compared to 89 L h 1 m 2 for NFand DK and 61 L h 1 m 2 for NF 245), rejection of sugars decreases which was already observed by Dalwani et al. [28] for NaCl in similar conditions and attributed to the occurrence of a polarization layer. Sugar rejection order is in accordance with MW and with Stokes diameter (0.73 for glucose and 0.63–0.64 for the C5 sugars). In spite of equivalent size, arabinose is always a little more rejected than xylose whatever the NF membrane used, which could be related to its higher hydration number (7.6 for arabinose compared to 6.8 for xylose, according to Galema and Hoiland [32]). Such result was also observed by Hua et al. [33]. Hydration also explains probably why sugars are rejected more than 83% even if their Stokes diameters are smaller than the reported membrane pore diameter (see Table 2). For xylose, higher rejection with DK N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 the retentate side, osmotic pressure difference ΔΠ increases, leading to a lower effective TMP and a lower permeate flux according to the solution–diffusion model:   J P ¼ AW U TMP ΔΠ ð6Þ where ∆П is the osmotic pressure difference between retentate and permeate (Pa). Experimental points were fitted with simple mathematical models for the sake of clarity. It can be observed that permeate flux decrease is linear with RO membranes, whereas it is logarithmic with NF membranes. Assuming Eq. (6) applies, linear behavior with RO can easily be related to the quite total rejection of sugars: in this case, ΔΠ increases proportionally to sugars concentration in the retentate (according to Van't Hoff law) and sugars concentration factor is equal to VRR. Then, JP varies linearly with VRR. 3.3.2. Effect on rejection with RO membranes With RO membranes, concentration increase does not change the rejection of the sugars that were already completely rejected at VRR ¼1. The same statement holds for vanillin (Fig. 5). For acetic acid and HMF, a slight decrease is observed but globally rejection does not change much (Fig. 5). Furfural presents a distinct behavior with a strong decrease especially for SG and CPA membranes (from 40% at VRR ¼1–4% at VRR ¼2 for CPA2). It is observed for acetic acid and HMF that the results obtained in the concentration mode fits perfectly with results obtained in the recycling mode (cf. Fig. 2). Provided permeate flux is known, rejection can be deduced independently of the experimental conditions (VRR). However, this is not true in the case of furfural, and for CPA and SG membranes: at equivalent permeate flux, 80 70 60 3.3. Effect of concentration on performances NF270 DK NF- NF245 NF90 ESPA2 XLE CPA2 CPA3 SG 50 -2 -1 compared to NF270 is surprising because pore diameters values are larger. Regarding inhibitors, NF membranes give very low rejection (high transmission) which is convenient for the detoxification purpose. NF270 and DK give the smallest rejections for all inhibitors, with plateau at 2%, 5%, 8% and 12% for furfural, acetic acid, HMF and vanillin, respectively. This order of rejection is identical to that already observed for most of the RO membranes (at the exception of XLE), namely furfuraloacetic acidoHMFovanillin. It is worth noting the significant difference of behavior for vanillin and C5 sugars: with similar MW, vanillin (MW¼ 152 g mol 1) transmission is 480%, whereas xylose and arabinose (MW¼150 g mol 1) rejection is 480%. With log KOW ¼ 1.21, vanillin has probably a stronger affinity than sugars for the aromatic polyamide material which results in a higher permeability and enhanced transport. Another explanation is that at equivalent MW, hydrophobic molecules are less hydrated than hydrophilic ones and have a smaller effective molecular size; they would therefore be less rejected ([8]). No results have been reported for vanillin but Maiti et al. [15] observed a high transmission of vanillic and ferulic acids through NF membranes of MWCO range equivalent to the actual ones, which was attributed to their low Stokes diameters: 0.48 and 0.58 nm compared to 0.64 and 0.73 for xylose and glucose, respectively (Table 1). Actually, high transmission of phenolic compounds is very promising because they were expected to be the most difficult to separate from sugars through pressure-driven membrane processes. Finally, Table 4 sums up rejection results at a permeate flux of about 18 L h 1 m 2 for RO and 65 L h 1 m 2 for nanofiltration membranes, corresponding to different TMP according to the membranes. Nanofiltration offers at higher flux and lower pressure a better detoxification effect of lignocellulosic hydrolysate model solution than reverse osmosis. NF- achieves the best rejection of sugars, but NF270 and DK membranes give the lowest rejection (best transmission) of inhibitors. With lower sugar loss, especially C5 sugars, DK should be preferred to NF270 although it should be operated at higher TMP to reach equivalent permeate flux (15 bars instead of 10). However, final conclusions on the choice of the membranes should take account of their performances when increasing sugars concentration by retentate recycling. Jp (L h m ) 46 Effect of concentration was studied with all membranes by extracting permeate and recycling retentate to the feed tank then increasing VRR up to 4 for RO and to 8 for NF. TMP was maintained constant at 30 bars for RO experiments and at 10 bars for NF, including NF90 which was considered with the other nanofiltration membranes in this part of the study. 40 30 20 10 0 0 2 4 6 8 VRR 3.3.1. Effect on permeate flux Permeate flux JP decreases when increasing VRR (Fig. 4). Indeed, when increasing concentration of the rejected species (sugars) on Fig. 4. Permeate flux versus VRR (NF: open symbols and dotted lines; RO: filled symbols and full lines. The curves correspond to mathematical fitting). (TMP is 10 bars for NF and 30 bars for RO; 20 1C; feed flow rate¼400 L h 1). Table 4 Rejection (%) of sugars and inhibitors in the recycling mode (VRR¼ 1). TMP (bar) Permeate flux Jp (L h Xylose Arabinose Glucose Acetic acid Furfural HMF Vanillin 1 m 2 ) SG CPA3 CPA2 XLE ESPA2 NF90 NF245 NF- DK NF270 30 18 98.4 97.7 98.8 60.5 50.7 84.1 92.0 20 20 20 15 10 15 15 10 97.5 96.9 97.8 40.1 28.1 66.5 86.7 97.4 96.7 97.3 43.5 31.4 31.4 89.2 98.5 97.4 98.8 79.8 87.6 98.0 94.3 98.8 98.0 98.8 54.5 47.0 85.8 93.9 100.0 100.0 100.0 50.0 42.1 81.6 84.0 20 65 89.8 91.3 95.5 6.7 2.3 12.7 19.0 91.6 93.7 97.3 7.8 0.5 12.8 13.3 90.1 93.3 96.6 4.9 –0.9 7.1 9.8 82.9 83.6 94.8 4.5 0.5 6.5 7.8 N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 47 XLE 100 R (%) 80 60 40 Vanillin HMF Acetic acid Furfural 20 0 0 1 2 3 4 5 VRR ESPA2 80 80 60 40 R (%) R (%) SG 100 100 Vanillin HMF Acetic acid Furfural 20 60 40 Vanillin HMF Acetic acid Furfural 20 0 0 0 1 2 3 4 5 0 1 2 VRR 5 CPA3 100 80 80 60 60 R (%) R (%) 4 VRR CPA2 100 3 40 Vanillin HMF Acetic acid Furfural 20 1 Vanillin HMF Acetic acid Furfural 20 0 0 0 40 2 3 4 0 5 VRR -20 1 2 3 4 5 VRR Fig. 5. Inhibitors rejection versus VRR with RO membranes (TMP¼ 30 bars; 20 1C; feed flow rate¼ 400 L h 1 ). furfural is far more transmitted in the concentration mode than in the recycling mode. This observation is difficult to explain on the sole basis of the present experiments. Furfural rejection decrease with VRR is interesting for the detoxification purpose; however, the other inhibitors are mostly unaffected. similar comment holds for vanillin. Regarding HMF, acetic acid and furfural, NF90 behaves more like the other NF membranes, with rejection decreasing when VRR increases; however, values achieved are still high compared to the others. Hybrid behaviour of NF90 is probably to relate to its active layer, the only one among the NF membranes to be made of fully aromatic polyamide like RO. 3.3.3. Effect on rejection with NF membranes With NF membranes (Fig. 6), sugar rejection decreases noticeably with VRR and this is more marked for xylose and arabinose (up to 14% between VRR ¼1 and VRR ¼ 8 for NF270 for example) than for glucose. Rejection of inhibitors also decreases, reaching even negative values for acetic acid, furfural and HMF. The decrease of rejections observed in NF is again directly related to the permeate flux decrease. Negative rejection values for acetic acid (in the presence of xylose) and for furan derivatives were reported by Weng et al. ([11,12]) and Qi et al. ([14]) who attributed enhanced transport of inhibitors to interactions with polarisation layer. NF90 presents a specific behaviour. Regarding sugars, it presents the same behaviour as the RO membranes, with constantly high sugar rejection and no difference between glucose and C5 sugars; 3.4. Choice of NF or RO membranes for detoxification Inhibitory compounds act differently on the fermentation yeasts and do not show the same toxicity levels. However, as commonly reported, the presence of several toxic molecules severely enhances the inhibitory effects. This would lead to prefer NF to RO for detoxification. At the highest tested VRR (VRR¼8) and TMP¼10 bars, nanofiltration membranes (except NF90) lead to very high transmission of inhibitors (496%), NF- and NF270 being especially remarkable with negative rejection values for acetic acid, furfural and HMF (Table 5). Other criteria are rejection of sugars that should be as high as possible to minimize sugar loss and permeate flux. NF- shows higher sugar rejection than DK, NF245 and NF270 (þ9% for xylose and arabinose and þ6% for glucose compared to NF270). However, NF270 gives the highest flow rate (20 L h 1 m 2) followed by DK, NF- and 48 N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 DK DK Vanillin HMF Acetic acid Furfural 10 5 100 R (%) R (%) 15 90 80 Glucose Xylose Arabinose 70 0 0 3 -5 6 60 9 0 NF 270 10 R (%) R (%) Vanillin HMF Acetic acid Furfural 5 0 80 Glucose Xylose Arabinose 60 6 9 0 3 6 9 VRR VRR -5 NF- NF- 10 100 R (%) Vanillin HMF Acetic acid Furfural 15 R (%) 90 70 3 9 100 15 0 6 VRR NF270 5 90 80 Glucose Xylose Arabinose 70 60 0 0 3 -5 6 0 9 3 6 9 VRR VRR NF245 NF245 100 Vanillin HMF Acetic acid Furfural 10 R (%) 15 R (%) 3 VRR 5 90 80 Glucose Xylose Arabinose 70 60 0 0 3 -5 6 0 9 3 VRR NF90 100 6 9 VRR NF90 100 90 R (%) R (%) 80 60 40 Vanillin HMF Acetic acid Furfural 20 0 0 80 Glucose 70 Arabinose Xylose 60 3 6 VRR 9 0 3 6 9 VRR Fig. 6. Sugars and inhibitors rejection versus VRR with NF membranes (TMP¼ 10 bars; 20 1C; feed flow rate¼ 400 L h 1 ). N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 49 Table 5 Rejection (%) of sugars and inhibitors at VRR¼ 4 for RO membranes and VRR¼ 8 for NF membranes. TMP (bar) VRR Permeate flux Jp (L h Xylose Arabinose Glucose Acetic acid Furfural HMF Vanillin 1 m 2 ) SG CPA3 CPA2 XLE ESPA2 NF90 NF245 NF- DK NF270 30 4 8.8 98.0 97.8 98.1 50.6 10.3 77.7 90.3 30 30 30 30 10 10 10 10 12.3 98.1 98.0 98.5 41.4 0.9 71.8 92.7 13.6 98.0 97.8 98.2 45.5 4.0 74.7 93.0 13.9 98.6 98.3 98.9 79.0 81.4 96.5 96.6 17.8 99.1 98.8 99.3 64.9 42.4 90.3 97.1 10 8 1.9 97.1 97.4 97.4 18.2 17.4 58.3 81.2 8.9 72.6 76.2 91.9 2.5 0.3 1.1 4.1 10.1 77.2 78.6 92.1 2.4 4.1 2.9 1.5 11.0 73.8 77.8 92.5 2.0 0.6 1.2 3.8 20.0 68.4 69.8 86.5 0.9 3.5 2.3 1.2 NF245 (11, 10 and 9 L h 1 m 2, respectively). Final decision should be made between NF270, DK and NF- after experiments at larger scale with spiral-wound configuration more representatives of industrial processing and with real hydrolysates. Increasing VRR will allow the concentration of sugars to increase more or less proportionally in the nanofiltrated hydrolysate which is beneficial for energy consumption at the distillation stage. However, even in the most favourable case of 100% transmission, inhibitors concentration will remain unchanged. To achieve the required detoxification effect, nanofiltration will therefore have to be operated in a diafiltration mode. Optimization of processing conditions will have to take into account fermentability assessment. 4. Conclusion Although they were selected in a rather narrow range of MWCO (100–300 g mol 1) and showed some similarities (pore diameter, water permeability, sugar rejection), tight NF and RO membranes differentiate significantly regarding inhibitors rejection. With somewhat larger MWCO, NF leads to a strong selectivity difference between C5 sugars and vanillin of similar MW due to differences in hydration and hydrophobicity. With high transmission of inhibitors (496% at VRR¼ 8), NF membranes seem more appropriate for detoxification purpose than RO although a loss of C5 sugars is visible at high VRR. Acknowledgments This work is part of Dr. N. Nguyen doctorate thesis. Vietnam International Education Development (VIED) is acknowledged for contributing financially to Dr. N. Nguyen stay in France. References [1] E. Palmqvist, B. Hahn-Hägerdal, Fermentation of lignocellulosic hydrolysates. II: Inhibitors and mechanisms of inhibition, Bioresour. Technol. 74 (2000) 25–33. [2] S. Mussatto, I. Roberto, Alternatives for detoxification of diluted-acid lignocellulosic hydrolyzates for use in fermentative processes: a review, Bioresour. Technol. 93 (2004) 1–10. [3] H.J. Huang, S. Ramaswamy, U.W. Tschirner, B.V. Ramarao, A review of separation technologies in current and future biorefineries, Sep. Purif. Technol. 62 (2008) 1–21. [4] P.T. Pienkos, M. Zhang, Role of pretreatment and conditioning processes on toxicity of lignocellulosic biomasse hydrolysates, Cellulose 16 (4) (2009) 743–762. [5] C. Abels, F. Carstensen, M. Wessling, Membrane processes in biorefinery applications, J. Membr. Sci. 444 (2013) 285–317. [6] M.A Franden, H.M. Pilath, A. Mohagheghi, P.T. Pienkos, M. Zhang, Inhibition of growth of Z. mobilis by model compounds found in lignocellulosic hydrolysates, Biotechnol. Biofuels 6 (2013) 99. [7] C. Bellona, J.E. Drewes, P. Xu, G. Amy, Factors affecting the rejection of organic solutes during NF/RO treatment: a literature review, Water Res. 38 (2004) 2795–2809. [8] B. Van der Bruggen, A. Verliefde, L. Braeken, E.R. Cornelissen, K. Moons, J. Verbeck, J.C. van Dijk, G. Amy, Assessment of a semi-quantitative method for estimation of the rejection of organic compounds in aqueous solution in nanofiltration, J. Chem. Technol. Biotechnol. 81 (2006) 1166–1176. [9] E. Drazevic, S. Bason, K. Kosutic, V. Freger, Enhanced partitioning and transport of phenolic micropollutants within polyamide composite membranes, Envir. Sci. Technol. 46 (2012) 3377–3383. [10] L.D. Nghiem, P.J. Coleman, NF/RO filtration of the hydrophobic ionogenic compound triclosan: transport mechanisms and influence of membrane fouling, Sep. Purif. Technol. 62 (2008) 709–716. [11] Y. Weng, H. Wei, T. Tsai, W. Chen, T. Wei, C. Hwang, C. Huang, Separation of acetic acid from xylose by nanofiltration, Sep. Purif. Technol. 67 (2009) 95–102. [12] Y. Weng, H. Wei, T. Tsai, T. Lin, T. Wei, G. Guo, C. Huang, Separation of furans and carboxylic acids from sugars in dilute acid rice straw hydrolyzates by nanofiltration, Bioresour. Technol. 101 (2010) 4889–4894. [13] F. Zhou, C. Wang, J. Wei, Separation of acetic acid from monosaccharides by NF and RO membranes: performance comparison, J. Membr. Sci. 429 (2013) 243–251. [14] B. Qi, J. Luo, X. Chen, X. Hang, Y. Wan, Separation of furfural from monosaccharides by nanofiltration, Bioresour. Technol. 102 (2011) 7111–7118. [15] S.K. Maiti, Y.L. Thuyavan, S. Singh, H.S. Oberoi, G.P. Agarwal, Modeling of the separation of inhibitory components from pretreated rice straw hydrolysate by nanofiltration membranes, Bioresour. Technol. 114 (2012) 419–427. [16] C. Sagne, C. Fargues, B. Broyart, M.-L. Lameloise, M. Decloux, Modeling permeation of volatile organic molecules through reverse osmosis spiralwound membranes, J. Membr. Sci. 330 (2009) 40–50. [17] C. Sagne, C. Fargues, R. Lewandowski, M.-L. Lameloise, M. Gavach, M. Decloux, A pilot scale study of reverse osmosis for the purification of condensate arising from distillery stillage concentration plant, Chem. Eng. Process. 49 (2010) 331–339. [18] N. Nguyen, Detoxification of Ligno-Cellulosic Hydrolyzates By Pressure-Driven Membrane Processes: Interest of Nanofiltration in a Diafiltration Mode, AgroParisTech, Massy, France, 2014 (Doctorate thesis). [19] A. Verliefde, N. Van Vliet, G. Amy, B. Van der Bruggen, J.C. van Dijk, A semiquantitative method for prediction of the rejection of uncharged organic micropollutants with nanofiltration, Water Pract. Technol. 1 (4) (2006) 1–10. http://dx.doi.org/10.2166/WPT.2006084. [20] C. Zidi, R. Tayeb, N. Boukhili, M. Dhahbi, A supported liquid membrane system for efficient extraction of vanillin from aqueous solutions, Sep. Purif. Technol. 82 (2011) 36–42. [21] Y. Bennani, K. Kosutic, E. Drazevic, M. Rozic, Wastewater from wood and pulp industry treated by combination of coagulation, adsorption on modified clinoptilotite tuff and membrane processes, Environ. Technol. 33 (10) (2012) 1159–1166. [22] L.D. Nghiem, S. Hawkes, Effects of membrane fouling on the nanofiltration of pharmaceutically active compounds (PhACs): mechanisms and role of membrane pore size, Sep. Purif. Technol. 57 (2007) 182–190. [23] E. Chilyumova, J. Thoming, Nanofiltration of bivalent nickel cations: model parameter determination and process simulation, Desalination 224 (2008) 12–17. [24] J. Warczok, M. Ferrando, F. Lopez, C. Guell, Concentration of apple and pear juices by nanofiltration at low pressures, J. Food Eng. 63 (2004) 63–70. [25] C. Sagne, C. Fargues, R. Lewandowski, M.-L. Lameloise, M. Decloux, Screening of reverse osmosis membranes for the treatment and reuse of distillery condensates into alcoholic fermentation, Desalination 219 (2008) 335–347. [26] H. Lonsdale, U. Merten, R. Riley, Transport properties of cellulose acetate osmotic membranes, J. Appl. Polym. Sci. 9 (1965) 1341–1362. [27] R. Gerard, H. Hachisuka, M. Hirose, New membrane developments expanding the horizon for the application of reverse osmosis technology, Desalination 119 (1998) 47–55. 50 N. Nguyen et al. / Journal of Membrane Science 487 (2015) 40–50 [28] M. Dalwani, N.E. Benes, G. Bargeman, D. Stamatialis, M. Wessling, Effect of pH on the performance of polyamide/polyacrylonitrile based thin film composite membranes, J. Membr. Sci. 372 (2011) 228–238. [29] C. Fargues, C. Sagne, A. Szymczyk, P. Fievet, M.-L. Lameloise, Adsorption of small organic solutes from beet distillery condensates on reverse-osmosis membranes: consequences on the process performances, J. Membr. Sci. 446 (2013) 132–144. [30] D. Violleau, H. Essis-Tome, H. Habarou, J.P. Croué, M. Pontié, Fouling studies of a polyamide nanofiltration membrane by selected natural organic matter: an analytical approach, Desalination 173 (2005) 223–238. [31] E. Sjöman, M. Mänttäri, M. Nyström, H. Koivikko, H. Heikkilä, Separation of xylose from glucose by nanofiltration from concentrated monosaccharide solutions, J. Membr. Sci. 292 (2007) 106–115. [32] S.A. Galema, H. Hoiland, Stereochemical aspects of hydration of carbohydrates in aqueous solutions. Density and Ultrasound measurements, J. Phys. Chem. 95 (1991) 5321–5326. [33] X. Hua, H. Zhao, R. Yang, W. Zhang, W. Zhao, Coupled model of extented Nernst–Planck equation and film theory in nanofiltration for xylooligosaccharide syrup, J. Food Eng. 100 (2010) 302–309.