Production of purified hemicellulose-pectin fractions of different molecular weight from discarded carrots by hydrothermal treatment followed by multistep ultrafiltration/diafiltration

Journal of Cleaner Production 321 (2021) 128923 Available online 9 September 2021 0959-6526/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Production of purified hemicellulose-pectin fractions of different molecular weight from discarded carrots by hydrothermal treatment followed by multistep ultrafiltration/diafiltration Marta Ramos-Andr´ es, Beatriz Aguilera-Torre, Juan García-Serna * Grupo de Ingeniería de Procesos a Presi´ on, Instituto de Bioeconomía de la Universidad de Valladolid (BioEcoUVa), Departamento de Ingeniería Química y Tecnología del Medio Ambiente, Escuela de Ingenierías Industriales, Universidad de Valladolid, 47011, Valladolid, Spain ARTICLE INFO Handling Editor: Zhen Leng Keywords: Hydrothermal Hemicellulose Pectin Polydispersity Membrane Film ABSTRACT Hemicelluloses and pectins are good candidates as biopolymers for the formation of products such as packaging films. Purified freeze-dried fractions of hemicelluloses and pectins, of different molecular weights, were obtained by treating hydrothermal extracts (140, 160, and 180 ◦C) of discarded carrots with ultrafiltration membranes (30, 10, 5, and 1 kDa). After each ultrafiltration, several cycles of diafiltration with partial water reuse were applied, obtaining a better separation and purification. A cascade configuration (30-10-5-1 kDa) was used in the 140 and 160 ◦C extracts, and a mixed configuration (5-10-1 kDa) in the 180 ◦C extract. High molecular weight hemicelluloses increased in concentration by a factor of 5 in the cascade configuration and by a factor of 16.67 in the mixed configuration. A high removal of free sugars (98.9–99.5 wt%) and by-products (94.4–99.2 wt%) through 1 kDa permeate and diafiltration waters was obtained. The system allowed moving from feeds with molecular weight, polydispersity, and purity in the ranges 9.02–18.83 kDa, 16.2–31.6, and 30.12–33.51 wt% to fractions with values in the ranges 2.59–102.75 kDa, 1.2–4.0, and 73.1–100 wt%. 1. Introduction The global economy is highly dependent on fossil sources for products and energy. Their long-term consumption is environmentally unsustainable due to their depletion and the impact and pollution problems associated with their use. It is important to find other organic sources allowing the development of processes and products in a cleaner, more sustainable and renewable way. 1.1. Food waste is a worldwide issue One of the main waste products worldwide is vegetable food waste. This waste is characterized by a high moisture content which makes it difficult to manage. Most of it is dumped into the field or landfills, while a small part is used as animal feed. This behavior creates environmental problems due to the putrefaction processes occurring quickly in the area. In particular, carrot production generates around 30 wt% of discards due to several reasons (Mari´ c et al., 2018): damage during the production process, not meeting market standards in terms of shape, size or colour, pulp generated in juice companies and withdrawal in supermarkets due to some degradation, among others. Although the priority is to minimize the waste, it is of great importance to be able to treat it appropriately to take advantage of the components of interest it contains. Discarded carrots are a source of vitamins, carotenoids, minerals, sugars and dietary fiber. In previous studies, we have focused on the valorization of discarded carrot pulp and discarded carrot juice. The valorization of the juice focused on sugars and nutrients through fermentation, and carotenoids through encapsulation and drying. The valorization of the pulp was based on hydrothermal treatment for the extraction of free sugars and high molecular weight hemicelluloses and pectins, leaving a residual pulp with abundant cellulose content (Ramos-Andr´ es et al., 2020). The present work aims at the conditioning of the hydrothermally extracted hemicelluloses and pectins for use in the production of high value-added products. 1.2. Hemicelluloses Hemicelluloses are the second most abundant biopolymers in plants after cellulose. If hemicelluloses are extracted cost-efficiently in their * Corresponding author. E-mail address: [email protected] (J. García-Serna). Contents lists available at ScienceDirect Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro https://doi.org/10.1016/j.jclepro.2021.128923 Received 3 June 2021; Received in revised form 24 August 2021; Accepted 2 September 2021 Journal of Cleaner Production 321 (2021) 128923 2 polymeric form they can be used in the formation of hydrogels, biodegradable barrier films for food packaging, paper additives, adhesives, thickeners, emulsifiers, stabilizers, and binders in the food, pharmaceutical, and cosmetic industries (Thuvander and J¨ onsson, 2020). The most common methods used for hemicelluloses extraction are alkaline, organosolv, acid, and hydrothermal treatments, among others. Hydrothermal treatment has important advantages as it uses only biomass and water as reagents, with no need for subsequent neutralization or desalting, and reduces corrosion in the equipment used. The species responsible for the autohydrolysis phenomenon in the hydrothermal treatment are the hydronium ions formed in the water at temperatures above 120 ◦C and the organic acids released during extraction. Hemicelluloses extracted by almost any method, and especially by hydrothermal treatment, are accompanied by a number of co-extracted compounds such as monosaccharides, organic acids, extractives, degradation compounds and certain lignin derivatives. Hemicelluloses require purification to be used in any application. The main purification methods are ultrafiltration (UF), ethanol precipitation, and chromatography. Although the latter two methods seem to give more homogeneous fractions, UF is the most widely used in recent research due to its significant operational advantages (Arkell et al., 2013a). Membrane operation is characterized by low chemical consumption (for cleaning only), low energy requirement (with operating conditions close to ambient), scalability and flexibility of operation, providing a cleaner and more efficient purification process. The main challenge of this purification method is membrane fouling, which is reduced by feed pretreatments such as microfiltration or some heating to reduce viscosity. Diafiltration (DF) is an important post-treatment to mitigate fouling. The reduction of the feed volume during UF allows concentration and partial purification, but DF ensures complete purification by removing unwanted substances. DF consists of diluting the retentate by adding water and applying UF, repeating this cycle until de desired level of impurities removal is reached. This method is probably the most cost-effective for hemicelluloses purification (Al Manasrah et al., 2012). 1.3. Pectins Pectins are heteropolysaccharides present in the primary cell wall and middle lamella of higher plants. They consist mostly of α -galactur- onic acid subunits with a variable content of methyl esters occurring in the acid group. This region of the pectins is called homogalacturonan. Other monosaccharides such as arabinose, galactose, and rhamnose may also be present in smaller proportions (Babbar et al., 2016). The industrial extraction of pectin is performed from fruit peel (waste from the juice industry), mainly from citrus fruits (Grassino et al., 2018). Other residues also contain pectins and although sometimes not in high quantities, they may be of better quality (Grassino et al., 2016). Some of these raw materials are olive pomace, berry pomace, potato pulp, and carrot pulp. Once extracted, pectins are used in the food industry as additives for thickening, gelling, stabilizing, and emulsifying agent in jams, jellies, soft drink, fish, meat, fruit juice, and milk product, as well as a fat replacer in spreads, salad dressing, ice cream, and emulsified meat products (Raji et al., 2017). For food applications, pectins are required to contain at least 65% galacturonic acid subunits (Schmidt et al., 2017). Another application starting to be studied is pectin in its oligosaccharide form, with a lower molecular weight, which can be used as a prebiotic agent (Babbar et al., 2016). The most common extraction method for pectins is acidic extraction using a dilute mineral acid (hydrochloric acid, nitric acid, and sulfuric acid). These acids have the advantage of low price and high extraction yield, but their major disadvantage is toxicity and harmful effects on the environment (Jafari et al., 2017). Other more environmentally friendly extraction methods are enzyme-assisted extraction, ultrasound-assisted extraction, microwave-assisted extraction, subcritical water extraction, and induced electric field-assisted extraction (Dranca and Oroian, 2018). After extraction, pectins have to be purified for application by reducing the presence of co-extracted sugars and acids. The main method is precipitation using an organic solvent, usually ethanol, either in a single step or in several steps. Other methods also used are dialysis, ionic exchange, nitration, as well as combined methods (Dranca and Oroian, 2018). Novel, more environmentally friendly methods include UF in combination with DF, UF, metal precipitation, copper precipitation, and purification using protein (sodium caseinate) (Dranca and Oroian, 2018). 1.4. Films from hydrothermally extracted biopolymers The hydrothermal extraction of hemicelluloses and pectins from discarded carrots by means of a flow-through configuration allows obtaining these biopolymers with high molecular weight, together with other components (Ramos-Andr´ es et al., 2020). A concentration, separation, and purification treatment as in the present work allows obtaining fractions suitable for film manufacture. Edible films are a good alternative for the conservation of food because they have good properties among which are their biodegradability and zero toxicity. They also stand out because they can act naturally as antimicrobials or antioxidants (Rojas et al., 2015). These films are characterized as being a very good barrier to gases, although not so good to water vapor, as their raw material (polysaccharides and/or proteins) is hydrophilic (Rojas et al., 2015). 1.5. Purification using membranes applying ultrafiltration and diafiltration Membranes have been used as a purification method with almost all types of hemicellulose extraction methods, such as hemicelluloses extracted by alkaline treatment and recovered using UF (Laine et al., 2015) and nanofiltration (Arkell et al., 2013b). Persson et al. (2010) fractionated process water from a thermomechanical pulp mill into five fractions using filtration, ultrafiltration, and nanofiltration. Rico et al. (2018) valorized peanut shells by hydrothermal extraction of hemicelluloses and lignin-derivative compounds. Successive DF stages allowed obtaining a purity of 72.4 wt%. Krawczyk et al. (2008) extracted hemicelluloses by steam explosion from barley husks. They applied prefiltration followed by an UF/DF system, which increased the purity of the hemicelluloses from 40 to 70 wt%. Several authors have used several membranes, either in series or in parallel, as in the present work. Gonz´ alez-Mu˜ noz et al. (2013) extracted hemicelluloses from Pinus pinaster wood with a purity of 69.6 wt%. Concentration, purification, and fractionation were carried out by successive stages of DF and concentration with MWCO 10, 5, 3, 1, and 0.3 kDa membranes. The purification effects were mainly seen in the 5 and 3 kDa membranes, enabling the generation of manooligosaccharides with tailored molecular weight distributions. In the case of Sukhbaatar et al. (2014), after extraction of sugarcane bagasse hemicelluloses by hydrothermal treatment, UF and nanofiltration were performed in series with MWCO 10, 1, and 0.4 kDa membranes to recover high molecular weight hemicelluloses and oligomers. Al Manasrah et al. (2012) applied hydrothermal treatment to spruce-sawdust and recovered galactoglucomannan which was further processed with three membranes of 5, 10, and 30 kDa. The best results were obtained with the 5 kDa membrane, with 88% retention, 63 wt% purity, and 70% recovery, using a feed volume reduction of 86 vol%. DF did not improve the purity of galactoglucomannan due to overlapping of the lignin molecular weight distribution. This problem was not present in this work due to the negligible lignin content in the extracts. In one of our previous works, hydrothermal treatment was applied to spent coffee grounds followed by cascade UF/DF with three membranes of MWCO 30, 10, and 5 kDa for separation and purification of the hemicelluloses (Ramos-Andr´ es et al., 2019). Thanks to DF, the retention of co-extracted by-products was reduced from 45.6 to 8.7 wt%, obtaining fractions with average molecular weights from 1.6 to 49.7 kDa M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 3 and purities between 83.7 and 97.8 wt%. Although these purity values were high, this was also because the purity of the feed was relatively high (75.2–84.3 wt%). In the present work, a higher number of DF cycles were applied as the purity values of the feeds were lower (approx. 30%) mainly due to free sugars. Compared to our previous work, in the present work an additional 1 kDa membrane was used to recover the abundant hemicelluloses between 1 and 5 kDa. Also, all the membranes used were bigger, which meant great advantages in the operation and allowed to obtain fractions of sufficient volume to proceed to drying and characterization of the solid. As in our previous work, the purity was high (73.1–100%) compared to the obtained in other studies (63.0–72.4%). Not only the high purity is remarkable, but also the molecular weight achieved in the present work was very high in relation to the usual due to both the flow-through extraction previously performed (Ramos-Andr´ es et al., 2020) and the efficient operation with the membranes. 1.6. Aim of this work To our knowledge, there are no studies using membranes for the recovery of components from discarded carrots, namely hemicelluloses, and pectins. Our previous work on hydrothermal treatment of discarded carrots at different temperatures yielded extracts with a very wide molecular weight distribution due to the presence of different hemicelluloses/pectins and especially the presence of free sugars (Ramos-Andr´ es et al., 2020). The present work is focused on the processing of these novel extracts characterized by excellent quality hemicelluloses/pectin of both very high and moderate molecular weight. The hydrothermal extracts of purity close to 30 wt% were processed with four cascade membranes applying UF, multiple DF cycles, and drying, which allowed obtaining defined and high-purity solid fractions of biopolymers containing hemicellulose and pectin. These fractions have different potential applications depending on their molecular weight. 2. Materials and methods 2.1. Raw material and sample preparation Discarded carrots were supplied by the company Mu˜ nozval S.L. (Valladolid, Spain). They were washed with tap water and stored in a cool, dry, and dark place for the shortest time. Discarded carrots were processed with a juice extractor (Moulinex ZU5008 Infinity Press Revolution) yielding approximately 0.545 kg of pulp per 1 kg of discards. While the juice was valorized following a different strategy, the pulp was subjected to hydrothermal treatment in a flow-through reactor for the extraction of free sugars, hemicelluloses, and pectins, as explained in our previous article (Ramos-Andr´ es et al., 2020). The biomass is loaded into the reactor with the help of a cartridge. The water is bypass through the plant until it reaches the desired temperature, after which it circulates through the heated reactor from top to bottom, performing the extraction. The reactor outlet stream exchanges heat with the inlet stream to save energy. Three experiments were performed extracting at 140, 160, and 180 ◦C for 80 min. The plant was kept pressurized to ensure the water was always in a liquid state, operating at 6 bar (140 ◦C), 10 bar (160 ◦C), and 15 bar (180 ◦C). The three hydrolysates obtained in the first 25 min of the three extractions were recovered to be subjected to a downstream process. The post-treatment was done immediately to prevent degradation, keeping the liquid at 4 ◦C whenever a stop in the process was necessary, for example in the cleaning of the membrane. The three hydrothermal extracts were characterized by determining their composition according to the method described in Section 2.4.1. and by determining their molecular weight distribution according to the method in Section 2.4.2. 2.2. UF and DF The hydrothermal extracts were subjected to several stages of UF and DF with membranes for the concentration, separation, and purification of hemicelluloses and pectins. Concentration was achieved through UF, separation through both UF and DF, and purification mainly through DF. Four UF membranes were used, three of them Pellicon 2 Mini Biomax polymeric membranes, flat sheet configuration, made of polyethersulfone, with molecular weight cut-off (MWCO) of 30, 10, and 5 kDa, and filtration area of 0.1 m 2 . The last membrane of 1 kDa was a Prep Scale-TFF polymeric membrane, spiral wound configuration, made of regenerated cellulose, and filtration area of 0.23 m 2 . To achieve the concentration effect, the feed volume was reduced by 80% (v/v) on each membrane during UF. To improve separation and to carry out purification, the retentates of each membrane were subjected to several DF cycles. In each cycle, a volume of water equal to the volume of the retentate was added and ultrafiltered being recovered as permeate. Each retentate was subjected to 5 DF cycles with distilled water, and up to 3 Fig. 1. A) Cascade and B) mixed configuration of the membranes including UF and DF cycles. M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 4 DF cycles with water reused from previous DFs according to Fig. 1. The transmembrane pressure was kept in the range 1.5–2.5 bar by a manually adjusted valve on the retentate side, and feed flow was approximately 4.8 L/m 2 /min. The first step was to subject each hydrothermal extract to dead-end microfiltration to remove particles larger than 50 μ m. After microfiltration, two membrane configurations were used: 1) Cascade configuration. It was applied to 140 and 160 ◦C hydrothermal extracts. The configuration was based on a cascade arrangement of the membranes following a decreasing order of MWCO: 30, 10, 5, and 1 kDa. The four retentates in the cascade underwent several cycles of DF as can be seen in Fig. 1a. The first retentate (Ret-30 kDa) underwent five cycles of DF, the second (Ret- 10 kDa) underwent six cycles, the third (Ret-5 kDa) underwent seven cycles, and the fourth retentate (Ret-1 kDa) underwent eight cycles. With Ret-30 kDa only distilled water was used for all five DF cycles. The remaining retentates were also subjected to five DF cycles with distilled water, but prior to these, the retentates were subjected to an additional cycle for each membrane located in the previous position in the cascade. These additional cycles were performed by reusing water from previous DF cycles to recover part of the entrained components. 2) Mixed configuration. The 180 ◦C extract was subjected to this configuration, using only three membranes, as GPC analysis indicates the absence of the highest molecular weight fraction. The first membrane was the 5 kDa MWCO, followed by the 10 kDa and 1 kDa membranes operating in parallel, each with one of the 5 kDa membrane outlets. DF cycles were carried out on the three retentates according to Fig. 1b. Ret-5 kDa underwent five DF cycles with distilled water, Ret-10 kDa also underwent five cycles with distilled water, and Ret-1 kDa, having a higher MWCO membrane before it, it underwent six DF cycles: the first cycle with reused water from the DF of Ret-5 kDa and the last five cycles with distilled water. The feed streams and all the products obtained in each membrane of MWCO ‘X’ were sampled: Ret-X kDa, Ret-X kDa-DF and Perm-X kDa. These liquid samples were analyzed for composition following the method described in Section 2.4.1. and for molecular weight distribution following the method described in Section 2.4.2. After each use, the membranes were subjected to a cleaning and flushing stage as recommended by the manufacturer. 2.3. Freeze-drying of the purified fractions Freeze-drying was used as drying method for the purified fractions of hemicelluloses and pectins obtained after UF/DF. With the cascade configuration using four membranes, four fractions were obtained: Ret- 30 kDa-DF, Ret-10 kDa-DF, Ret-5 kDa-DF, and Ret-1 kDa-DF. With the mixed configuration using three membranes, three purified fractions were obtained: Ret-10 kDa-DF, Perm-10 kDa, and Ret-1 kDa-DF. These liquid products were frozen overnight at −25 ◦C and then freeze-dried under vacuum (0.180 mbar) for 96 h using a Telstar Lyoquest −55 unit (Terrassa, Spain). 2.4. Analysis 2.4.1. Chemical characterization Composition analysis of the liquid samples was carried out by High- Performance Liquid Chromatography (HPLC) as previously described by Ramos-Andr´ es et al. (2019). A column SUGAR SH-1011 Shodex was used for the analysis of sugars, uronic acids, aldehydes, organic acids, and degradation compounds. The temperature in the column was maintained at 50 ◦C and the mobile phase had a flow of 0.8 mL/min of 0.01 N sulfuric acid in Milli-Q water. Sugars, uronic acids, aldehydes, and organic acids were identified and quantified with a Waters IR detector 2414. Degradation compounds (5-HMF and furfural) were analyzed with a Waters dual λ absorbance detector 2487 (210 nm and 254 nm). To determine the concentration of hemicelluloses and pectins, liquid samples were submitted to standardized hydrolysis to break the hemicelluloses and pectins into monomers. Briefly, 15 mL of Milli-Q water and 0.8 mL of sulfuric acid (72 wt%) were mixed with 5 mL of the sample. The solution was autoclaved at 121 ◦C for 1 h, and then neutralized with CaCO 3 and filtered (Pore size 0.22 μ m, Nylon; FILTER-LAB) before HPLC analysis. The standards used for the analysis were: sucrose (99%), galacturonic acid (97%), glucose (99%), galactose (99%), fructose (99%), arabinose (99%), glycolaldehyde (99%), formic acid (98%), acetic acid (98%), levulinic acid (98%), acrylic acid (99%), 5-hydroxymethylfurfural (99%), and furfural (99%), all of them from Sigma-Aldrich. The concentration of arabinogalactan was calculated using anhydrous corrections of 0.9 (galactose) and 0.88 (arabinose). Galacturonic acid from pectins was corrected by 0.9 as an anhydrous factor. 2.4.2. Molecular size distribution Molecular size distribution was determined by Size Exclusion Chromatography (HPLC-SEC) using the method previously described by Ramos-Andr´ es et al. (2019). A GPC column (SB-803 HQ; Shodex) was used together with a guard column (SB-G; Shodex). The temperature in the column was maintained at 35 ◦C. The mobile phase had a flow rate of 0.5 mL/min (NaNO 3 0.1 M +NaN 3 0.02% in Milli-Q water). A Waters IR detector 2414 was used to determine the molecular weight distribution of the liquid samples. The calibration curve was determined with a set of 5 pullulan standards (STANDARD P-82; Shodex) dissolved in Milli-Q water, ranged between 6.1 and 113 kDa of weight-average molecular weight. The chromatographic curves were integrated and the area under the curve was divided into regions for classifying in different groups depending on the molecular weight: (1) monomers, (2) dimers, (3) pentamers, (4) hexamers, (5) 1–5 kDa, (6) 5–10 kDa, (7) 10–30 kDa, and (8) >30 kDa. The groups of trimers and tetramers were under the limit of detection. 2.4.3. Structural characterization (FTIR, TGA) The purified solid fractions of hemicelluloses and pectins were structurally characterized by Attenuated total reflectance (ATR) – Fourier transform infrared spectroscopy (FT-IR) (Bruker, Alpha model, with a Platinum ATR single reflection diamond module). The absorbance spectra were obtained in a wavenumber range from 4,000 to 400 cm −1 , acquiring 64 scans per sample at a resolution of 4 cm −1 . The structure of each purified fraction was also studied through thermogravimetric analysis (TGA) carried out in a TGA/SDTA RSI analyzer of Mettler Toledo. Samples of approximately 10 mg were heated at 10 ◦C/min under nitrogen atmosphere (60 NmL/min) from a temperature of 50 ◦C up to around 900 ◦C. 3. Results and discussion The main data of the paper used in the tables and figures can be accessed here. 3.1. Hydrothermal extracts characterization It was analyzed the composition of the three hydrothermal extracts recovered in the first 25 min of the flow-through extraction, obtained at 140 ◦C (Feed-140), 160 ◦C (Feed-160), and 180 ◦C (Feed-180). These extracts were the feeds of the UF/DF system and their composition is listed in Table 1. As in the present work, a previous work has shown extracted hemicelluloses from carrot pulp were of the arabinogalactan type, with a majority proportion of galactose versus arabinose (Ribas-Agustí et al., 2014). As shown in Table 1, the concentration of arabinogalactan was 592.7 mg/L (140 ◦C), 935.3 mg/L (160 ◦C), and 1,556.2 mg/L (180 ◦C) M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 5 in the hydrothermal extracts obtained. Pectin biopolymer of the homogalacturonan type was present with a concentration of 289.56, 492.2, and 617.5 mg/L. Despite the biopolymers degradation caused by a higher extraction temperature, the extraction kinetics was accelerated along with temperature resulting in a higher concentration of both biopolymers in the collected extracts corresponding to the first 25 min of extraction time. The free sugars were the majority components of the extracts. These are the sugars also present in the juice, the most abundant being sucrose followed by glucose and fructose. Again, despite the degradation, a higher concentration of free sugars was obtained at a higher temperature during the first 25 min of extraction: 1,765.41 mg/L (140 ◦C), 2,340.79 mg/L (160 ◦C), and 3,093.90 mg/L (180 ◦C). By-products accompanying hemicelluloses, pectins, and free sugars can be classified into those of the first degree coming from the transformation of one of the products (glycolaldehyde, 5-HMF, furfural, arabinose, and acetic acid) and those of the second degree coming from the transformation of other by-products (levulinic acid, formic acid, and acrylic acid) (Ramos-Andr´ es et al., 2020). The total concentration of by-products was 175.45 mg/L (Feed-140), 491.72 mg/L (Feed-160), and 952.82 mg/L (Feed-180). Fig. 2 shows the molecular weight distribution of the three hydrothermal extracts. In this distribution, the main molecular groups can be identified. The first peak on the left in the chromatograms is associated with the monomers (glucose, fructose, and arabinose) and the second with the dimers (sucrose). Other important groups were close to 1 kDa, between 1 and 5 kDa, between 5 and 10 kDa, and >30 kDa. The use of UF membranes with MWCO of 1, 5, 10, and 30 kDa allows separating these groups and obtaining more defined fractions with less polydispersity. The polydispersity of the hydrothermal extracts was too high to allow specific application as biopolymers. The main cause of this high value was the presence of free sugars together with the extraction of hemicelluloses and pectins of the highest molecular weight. The weightaverage molecular weight (MW) and PD parameters of the molecular weight distributions shown in Fig. 2 were: 18,833 Da and 31.55 (Feed- 140), 10,977 Da, and 19.94 (Feed-160), 9,017 Da, and 16.16 (Feed- 180). As it can be seen in Table 1 and Fig. 2, Feed-180 had a higher presence of monomers than Feed-140 and Feed-160, which is associated with a greater extraction, greater hydrolysis of sucrose into its monomers, and a greater autohydrolysis of arabinogalactan releasing arabinose units. As for the molecular weight groups higher than sucrose but < 30 kDa, Feed-180 was the most concentrated feed, being Feed-140 and Feed-160 similar to each other. This higher concentration can be associated with both a higher extraction and the phenomenon of autohydrolysis at a higher temperature. With respect to the molecular group > 30 kDa, the concentration was lower the higher the extraction temperature. This can be associated with the phenomenon of autohydrolysis during extraction. The autohydrolysis effect occur at temperatures higher than 120 ◦C when the water is ionized forming H 3 O + ions which cause partial depolymerization of hemicelluloses and pectins. The acetyl and methyl groups of hemicelluloses and pectins are cleaved leading to the subsequent formation of H 3 O + ions that autocatalyze the depolymerization process (Ramos-Andr´ es et al., 2020). 3.2. Concentration applying UF and DF The feed volume reduction in the UF process allowed to increase the concentration of each molecular group. Those groups whose molecular weight is higher than the MWCO of the membrane are expected to be concentrated since they are retained. Fig. 3 shows on the y-axis the socalled concentration factor, which represents the increase in concentration compared to the concentration in the feed stream of the membrane system (Feed-140, Feed-160, and Feed-180). The concentration factor of each molecular group was calculated as the ratio between the areas under the curve of the HP-SEC chromatogram corresponding to each molecular group, comparing the different streams of the UF/DF system with the feed stream. The extracts obtained at 140 and 160 ◦C were subjected to the cascade membrane configuration, while the 180 ◦C extract was processed with a mixed configuration (Fig. 1). We used these two configurations for optimization purposes, as the higher temperature did not provide very long biopolymers and the use of 30 kDa was dispensable. The different configurations resulted in differences in the concentration factor. The feed volume reduction was in all cases 80 v/v% so the maximum theoretical increase in concentration for those groups totally retained was 5-fold. As can be seen in Fig. 3A and B, the group >30 kDa reached a concentration factor of 5 in Ret-30 kDa. After DF, this factor was maintained in Ret-30 kDa-DF, so hemicelluloses and pectins >30 kDa were effectively retained. With regard to the molecular groups <30 kDa, they were partially retained in Ret-30 kDa, in greater proportion as their molecular weight increased. The concentration factors of these groups were found to be between 0.92 and 5.00 for Feed-140 and between 0.05 and 4.91 for Feed-160. After applying the five cycles of DF and obtaining Ret-30 kDa-DF, the retention of the groups <30 kDa decreased considerably. The concentration factor was 0 for molecular weight groups <1 kDa, and for groups >1 kDa it was between 0.20 and 4.55 (Feed-140) and between 0.70 and 3.77 (Feed-160). With regards to the retentates of the other membranes in the cascade Table 1 Chemical composition of the hydrothermal extracts extracted at 140 ◦C (Feed- 140), 160 ◦C (Feed-160), and 180 ◦C (Feed-180) feeding the membrane configurations. Feed-140 Feed-160 Feed-180 Biopolymers (mg/L) 882.3 1,427.5 2,173.7 Galacto- (mg/L) 356.1 660.4 1,223.1 Arabino- (mg/L) 236.6 274.9 333.1 Arabinogalactan (mg/L) 592.7 935.3 1,556.2 Pectin (mg/L) 289.6 492.2 617.5 Free sugars (mg/L) 1,765.4 2,340.8 3,093.9 Sucrose (mg/L) 820.1 1,012.9 1,166.6 Glucose (mg/L) 483.4 704.6 1,130.6 Fructose (mg/L) 461.9 623.3 796.7 By-products (mg/L) 175.5 491.72 952.8 Arabinose (mg/L) 21.88 n.d. 174.7 Glycolaldehyde (mg/L) 32.21 82.72 140.7 Formic Acid (mg/L) 30.74 216.1 278.6 Acetic Acid (mg/L) 41.58 45.40 88.32 Levulinic Acid (mg/L) 1.93 15.76 3.89 Acrylic Acid (mg/L) 45.31 120.6 218.1 5-HMF (mg/L) 1.80 11.09 45.54 Furfural (mg/L) n.d. n.d. 3.04 Fig. 2. Molecular weight distribution of the hydrothermal extracts extracted at 140 ◦C (Feed-140), 160 ◦C (Feed-160), and 180 ◦C (Feed-180) feeding the membrane configurations. M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 6 (Ret-10 kDa, Ret-5 kDa, and Ret-1 kDa) their concentration with respect to the feed did not increase as significantly since the great part of the extract was retained in the first membrane (Ret-30 kDa). The molecular groups in the three retentates showed a concentration factor around 1 when operated with Feed-140. When Feed-160 was processed, the concentration factor in Ret-10 kDa was less than 1, being concentrated mainly the pentamers and hexamers, probably the most abundant groups in the feed of the 10 kDa membrane. It is worth noting an increase of 2.32-fold in the concentration of Ret-1 kDa dimer group of Feed-160, which may be due to a not very effective mass transfer through the membrane in this case. After applying the multiple DF cycles, the overall concentration of Ret-10 kDa-DF, Ret-5 kDa-DF, and Ret- 1 kDa-DF was much lower than the feed concentration, but a more sharpen molecular weight distribution can be seen (Fig. 3A and B). Those molecular groups not interested in retaining (<1 kDa) experienced a maximum concentration factor of 0.05 for Feed-140 and 0.13 for Feed-160, both in Ret-1 kDa-DF, so the concentration of non-target groups was reduced considerably by DF. As for the target groups, on Ret-10 kDa-DF the group increasing its concentration the most was 5–10 kDa (as 10–30 kDa was very low in feed), on Ret-5 kDa-DF it was the corresponding group 5–10 kDa and on Ret-1 kDa-DF it was indeed the group 1–5 kDa. Feed-180 was subjected to the mixed membrane configuration, resulting in 3 products instead of 4 products: Ret-10 kDa-DF, Perm-10 kDa, and Ret-1 kDa-DF. With the first of the membranes (MWCO 5 kDa) the concentration of all the molecular groups increased considerably in Ret-5 kDa, especially those with a molecular weight >5 kDa, which experienced a concentration factor between 4.84 and 5 (Fig. 3C). Monomers and dimers experienced almost total retention in Ret-5 kDa, with concentration factors of 4.53 and 4.50. This can be associated with poor mass transfer due to the presence of high molecular weight groups in a membrane with a moderate value of MWCO. After applying the five DF cycles, the presence of monomers and dimers was totally reduced in Ret-5 kDa-DF, keeping the concentration of the rest of the groups the same. This DF process was effective for the purification of the biopolymers, but not for the separation of molecular groups. Ret-5 kDa-DF was passed through the 10 kDa membrane and gave rise to Ret-10 kDa in which groups with a molecular weight >5 kDa were concentrated with concentration factors between 15.47 and 25, being 25 the maximum possible given the feed volume reductions. After applying DF to Ret-10 kDa, a product with improved characteristics was obtained (Ret-10 kDa- DF). The interest groups, those >10 kDa, increased their concentration between 10.20 and 16.67. The non-interest groups present were exclusively the group 5–10 kDa with concentration factor 2.42 and the group 1–5 kDa with concentration factor 0.26. The second product was Perm- 10 kDa, which had a very low impurities content due to the DF applied on the 5 kDa membrane. The most concentrated group was 5–10 kDa, as it was the most abundant group in the feed. The third product, Ret-1 kDa-DF, was much diluted compared to the feed. It presented good characteristics for not having impurities associated with monomers and dimers, not having molecular groups of very high molecular weight, and having its group of interest (1–5 kDa) the highest concentration factor of the groups present. 3.3. Purification applying UF and DF Table 2 analyses the separation of each of the four groups of extracted components: arabinogalactan, pectins, free sugars, and by- Fig. 3. Concentration factor (CF) of the different groups of hemicelluloses-pectins in each of the UF and UF/DF products compared to the feed stream. M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 7 products (percentage is wt%). Each of the groups was distributed in the UF streams or products and in the UF/DF products. In the cascade configuration (Feed-140 and Feed-160) the UF products were Ret-30 kDa, Ret-10 kDa, Ret-5 kDa, Ret-1 kDa, and Perm-1 kDa, whose retentions together added 100 wt%. In the mixed configuration (Feed- 180), the UF products were Ret-10 kDa, Perm-10 kDa, Ret-1 kDa, and Perm-1 kDa, which together with DF water from 5 kDa membrane added 100 wt% of retention. After purifying through DF cycles, the products in the cascade configuration were Ret-30 kDa-DF, Ret-10 kDa-DF, Ret-5 kDa-DF, Ret-1 kDa-DF, and Perm-1 kDa, which together with the DF waters added up to 100 wt%. In the mixed configuration, the products of the DF were Ret-10 kDa-DF, Perm-10 kDa, Perm-1 kDa, and the total DF waters (5, 10, and 1 kDa), adding up to 100 wt% retention. 3.3.1. Arabinogalactan (AG) Regarding the arabinogalactan present in Feed-140, Table 2 shows how it was distributed mainly in Ret-30 kDa (59.31 wt%) and Perm-1 kDa (20.54 wt%), the products of higher and lower molecular weight. Part was retained in the intermediate retentates (Ret-10 kDa, Ret-5 kDa, and Ret-1 kDa), with values in the range 4.46–9.95 wt%. After applying DF, Ret-30 kDa-DF slightly reduced its retention (52.91 wt%). Nothing or a very small percentage of arabinogalactan was recovered in the diafiltered intermediate retentates, having Ret-10 kDa-DF, Ret-5 kDa- DF, and Ret-1 kDa-DF retentions lower than 1.89 wt%. The reduction in the retention of arabinogalactan taking place after DF could be partly compensated by a recovery of arabinogalactan thanks to the reuse of the DF water. Despite this, 23.40 wt% of the total arabinogalactan was lost through DF. A percentage close to 44 wt% was lost due to low molecular weight through Perm-1 kDa and the DF water, while 52.91 wt% was recovered as purified high molecular weight arabinogalactan in Ret-30 kDa-DF. In the case of Feed-160, most of the arabinogalactan was retained in the first membrane of the cascade (68.43 wt%), with the next largest part being removed through Perm-1 kDa (14.30 wt%). The remaining arabinogalactan was homogeneously retained in the intermediate retentates (Ret-10 kDa, Ret-5 kDa, and Ret-1 kDa) with values in the range 5.60–5.90 wt%. After applying the DF cycles the retention of arabinogalactan decreased in all membranes, as in the case of Feed-140. Although some of the removed arabinogalactan could be recovered by reusing the DF water, most of it was lost. The retentions decreased to 59.20 wt% (Ret-30 kDa-DF) and to the 1.40–3.65 wt% range (Ret-10 kDa-DF, Ret-5 kDa-DF and Ret-1 kDa-DF). A total of 34.16 wt% was eliminated through Perm-1 kDa and the DF waters, similar to the treatment of Feed-140. The mixed operation with Feed-180 resulted in a loss of arabinogalactan through Perm-1 kDa of 15.27 wt% while the loss in the Ret-5 kDa DF was nil, as only by-products were removed. A 55.30 wt% of the arabinogalactan was retained in Ret-10 kDa, while Perm-10 kDa contained 20.21 wt%. After applying DF to Ret-10 kDa and Ret-1 kDa, both reduced their arabinogalactan retention despite the reuse of DF water in Ret-1 kDa. Ret-1 kDa-DF retained 2.47% and Ret-10 kDa-DF a moderate 37.46 wt%. The high extraction temperature resulted in a loss of low molecular weight arabinogalactan through Perm-1 kDa and DF waters of 39.86 wt%. This is similar to the value obtained for Feed-140 and Feed- 160, so part of the autohydrolysated hemicelluloses was not lost but remained in the intermediate molecular weight product Perm-10 kDa. 3.3.2. Pectins In the case of pectins, unlike arabinogalactan, all of them were distributed in the retentates for Feed-140. 94.77 wt% were recovered in Ret-30 kDa, while the rest were recovered in the following two retentates: Ret-10 kDa (1.65 wt%) and Ret-5 kDa (3.58 wt%). Pectins can be identified as high molecular weight components or at least associated with high molecular weight arabinogalactan, as already demonstrated in our previous work (Ramos-Andr´ es et al., 2020). Once the DF cycles were performed, the retention in Ret-30 kDa-DF was 74.92 wt%. Thanks to Table 2 Recovery of arabinogalactan (AG), pectins, free sugars, and by-products after UF and UF/DF. 140 ◦C UF products UF/DF products Ret-30 kDa Ret-10 kDa Ret-5 kDa Ret-1 kDa Perm-1 kDa Ret-30 kDa-DF Ret-10 kDa-DF Ret-5 kDa-DF Ret-1 kDa-DF DF water AG 59.31% 5.74% 9.95% 4.46% 20.54% 52.91% n.d. 1.89% 1.26% 23.40% Pectins 94.77% 1.65% 3.58% n.d. n.d. 74.92% 9.15% 10.45% 5.48% n.d. Free sugars 6.56% 15.53% 14.55% 14.38% 48.98% n.d. n.d. n.d. 0.53% 50.49% Byproducts 33.16% 23.00% 21.43% 2.14% 20.27% 3.71% n.d. n.d. n.d. 76.02% 160 ◦C UF products DF products Ret-30 kDa Ret-10 kDa Ret-5 kDa Ret-1 kDa Perm-1 kDa Ret-30 kDa-DF Ret-10 kDa-DF Ret-5 kDa-DF Ret-1 kDa-DF DF water AG 68.43% 5.90% 5.60% 5.77% 14.30% 59.20% 1.40% 1.59% 3.65% 19.86% Pectins 71.08% 12.98% 7.03% 7.60% 1.31% 71.07% 9.54% 6.77% 6.58% 4.73% Free sugars 7.82% 18.44% 16.19% 14.89% 42.66% 0.59% n.d. 0.16% 0.32% 56.27% Byproducts 19.80% 14.46% 12.64% 10.84% 42.26% n.d. n.d. 0.75% 0.01% 56.98% 180 ◦C UF products DF products Ret-10 kDa Perm-10 kDa Ret-1 kDa DF 5 kDa Perm-1 kDa Ret-10 kDa-DF Perm-10 kDa Ret-1 kDa-DF DF water AG 55.30% 20.21% 9.22% n.d. 15.27% 37.46% 20.21% 2.47% 24.59% Pectins 55.61% 34.43% n.d. 9.96% n.d. 24.97% 38.24% 6.46% 30.33% Free sugars 0.82% 0.62% 18.21% 23.82% 56.53% n.d. 0.81% 0.08% 42.58% Byproducts 1.47% 4.94% 18.17% 9.18% 66.24% 0.18% 5.44% 0.03% 28.11% M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 8 the reuse of the DF water from the 30 kDa membrane, pectin retention was increased in the next membranes: Ret-10 kDa-DF (9.15 wt%), Ret-5 kDa-DF (10.45 wt%), and Ret-1 kDa-DF (5.48 wt%). The loss of pectins in the DF water and Perm-1 kDa was nil, so the recovery process was fully efficient. As can be seen in Table 3, products with pectins as major components and in different molecular weights were obtained, as pectins were present in higher proportion than arabinogalactan in Ret-10 kDa-DF, Ret-5 kDa-DF, and Ret-1 kDa-DF. Ret-30 kDa-DF, the product with the highest molecular weight, was more abundant in arabinogalactan. Due to the higher extraction temperature, the pectins had a lower molecular weight at 160 than at 140 ◦C, with a lower percentage retained in the first membrane (71.08 wt%). While in the case of Feed- 140 no pectin was lost through Perm-1 kDa, in the case of Feed-160 1.31 wt% were removed through this stream. In the other membranes, pectins were retained in percentages of 12.98 wt% (Ret-10 kDa), 7.03 wt% (Ret-5 kDa) and 7.60 wt% (Ret-1 kDa). The application of DF cycles kept the retention in Ret-30 kDa intact, so all pectins present there are either high molecular weight or bound to high molecular weight arabinogalactan. Retention in the next membranes was slightly reduced after DF, being in the range of 9.54–6.58 wt%. Compared to a zero-pectin loss in the case of Feed-140, in this case, the loss was a total of 6.04 wt%. According to the results very few pectins reach molecular weights of less than 1 kDa under these conditions. Due to the high extraction temperature, Feed-180 had the lowest proportion of high molecular weight pectins, resulting in a recovery of 55.61 wt% at Ret-10 kDa. Certain pectins (9.96 wt%) were removed through DF of the 5 kDa membrane, the first membrane in the configuration. After the application of the DF cycles to Ret-10 kDa and Ret-1 kDa, a large part of the pectins retained as high molecular weight pectins were carried over in the DF water. Part of the pectins washed out in the 5 kDa DF water were recovered in the 1 kDa membrane DF, passing from 0 to 6.46 wt% in Ret-1 kDa-DF. The final retention of the other two products was 24.97 wt% in Ret-10 kDa-DF and 38.24 wt% in Perm-10 kDa. In this case, most of the pectins had an intermediate molecular weight. The total loss of pectins was 30.33 wt%, again lower than the loss of arabinogalactan but significantly higher than the pectins loss in Feed-140 and Feed-160. 3.3.3. Free sugars Free sugars were not the target compounds in the present work, so they are intended to be poorly retained in the membranes. After performing UF of Feed-140, the retention of free sugars was 6.56 wt% in Ret-30 kDa and was in the range 14.38–15.53 wt% for Ret-10 kDa, Ret-5 kDa, and Ret-1 kDa. A total of 48.98 wt% was removed via Perm-1 kDa. This result shows the need for several cycles of DF to purify the biopolymers. After these cycles, the retention of free sugars became zero except for Ret-1 kDa-DF which retained a low 0.53 wt%. 50.49 wt% of the sugars were removed through the DF water. The total removal through Perm-1 kDa and DF waters was 99.47 wt%, very effective. The retention of free sugars of Feed-160 was almost identical to the case of Feed-140, as can be seen in Table 2. The removal of free sugars via Perm-1 kDa was 42.66% versus 48.98% in the case of Feed-140. The total removal through Perm-1 and DF waters was 98.93%, resulting again in a very efficient purification process. In the mixed configuration of Feed-180, very few sugars remained in the 10 kDa membrane products, as the previous 5 kDa membrane underwent DF cycles where 23.82 wt% of the sugars were removed along with 56.53 wt% sent to Perm-1 kDa via Perm-5 kDa. The highest retention percentage occurred at Ret-1 kDa with 18.21 wt%. After the application of the DF cycles, the retention in the retentates was practically nil. Perm-10 kDa maintained a low 0.81 wt%. Sugars removed were 42.58 wt% through the DF waters. The total removed through Perm-1 kDa and the DF waters was 99.11 wt%, again very efficient even in the mixed rather than cascade configuration. 3.3.4. By-products By-products are, together with free sugars, the components to be reduced in the treatment. In the case of Feed-140, by-products experienced higher retention in the membranes than free sugars, perhaps due to a higher interaction with the high molecular weight hemicelluloses and pectins. Only 20.27 wt% was removed by Perm-1 kDa, making DF cycles essential for purification. After these, only Ret-30 kDa-DF retained 3.71 wt% of the by-products, the remainder being removed through the DF waters. 96.29 wt% was removed through Perm-1 kDa and the DF waters. The by-products experienced lower retention on the membranes in the case of Feed-160 than in the case of Feed-140, which may be due to the lower molecular weight of Feed-160 leading to less interaction with the by-products. By-products removal through Perm-1 kDa was 42.26 wt % in this case compared to 20.27 wt% in Feed-140. After application of the DF cycles, by-products retention was reduced to zero except for a small percentage in Ret-5 kDa-DF (0.75 wt%) and Ret-1 kDa-DF (0.01 wt %). The total removal through Perm-1 and DF waters was 99.24 wt%. In the mixed configuration with Feed-180, a considerable proportion of by-products was removed via Perm-5 kDa and from there via Perm-1 kDa with a percentage of 66.24 wt%. In the 5 kDa DF, 9.18 wt% was removed, resulting in very low retention of by-products in the 10 kDa membrane. Ret-1 kDa was the product more abundant in by-products (18.17 wt%). After applying the DF cycles, the retentates were almost completely purified, with 28.11 wt% of the by-products being removed through the DF waters and a total of 94.35 wt% being removed through Perm-1 kDa and DF. Both cascade and mixed configurations are suitable for purification based on the removal of free sugars and by-products. The loss of hemicelluloses and pectins was similar in both configurations, and the small differences can be attributed to the different extraction temperatures. An important difference between the two configurations is the number of products, which is lower in the mixed configuration as fewer membranes are used. Given the low proportion of biopolymers in the 10–30 kDa group, it may be more interesting to use fewer membranes. Due to the low proportion of hemicelluloses and pectins of intermediate molecular weight, a high percentage was not recovered in the reuse of the DF water. A mixed configuration seems more efficient with respect to resources (membranes, time, water) in view of the products obtained. 3.4. Separation applying UF and DF It is possible to study the effect of separation not only in terms of individual components but also in terms of molecular groups. In Section 3.2. it was shown how each molecular group was more or less concentrated in the different retentates, and how DF decreased the concentration of the non-target groups. DF is key in the separation of molecular groups as well as in the purification of biopolymers. Fig. 4 shows the effect of separation due to UF but mainly due to DF showing the composition (wt%) of each molecular group in the feed and in the final M. Ramos-Andr´ es et al. Journal of Cleaner Production 321 (2021) 128923 9 diafiltered products. A large difference can be seen in the molecular weight distribution of each product before and after the UF and DF cycles. In general, the presence of non-target components (<1 kDa) is practically null in all the products obtained. The maximum percentage reached was in the case of Ret-1 kDa-DF from Feed-180 where the hexamers group reached 2.8 wt%. In the cascade configuration with Feed-140 and Feed-160, Ret-30 kDa-DF products were rich in the >30 kDa group with percentages of 66.2 wt% (140 ◦C) and 58.4 wt % (160 ◦C), as it can be seen in Fig. 4. These percentages were considerably higher than the percentage of > 30 kDa in the feeds, which was 12.7 wt% (140 ◦C) and 8.9 wt% (160 ◦C). The other molecular groups >1 kDa were also present in Ret-30 kDa-DF in decreasing order according to their decreasing molecular weight. The Ret-10 kDa-DF product was rich in the 5–10 kDa group for 140 ◦C (40.8 wt%) and in the 10–30 kDa target group for 160 ◦C (47.2 wt%). The higher presence of 5–10 kDa than 10–30 kDa in Ret-10 kDa-DF at 140 ◦C could be due to the almost total retention of the 10–30 kDa group in the 30 kDa membrane even after applying DF (Fig. 3), not allowing the recovery of part of the 10–30 kDa group in the reuse of water in 10 kDa DF. The Ret-5 kDa-DF product of 140 ◦C contained the 1–5 kDa group as the most abundant (56.1 wt%) and the 5–10 kDa target group in the case of 160 ◦C (50.7 wt%). Again, the 5–10 kDa was considerably concentrated in Ret-30 kDa-DF at 140 ◦C compared to 160 ◦C and despite DF (Fig. 3). The last of the products, Ret-1 kDa-DF, was rich in both cases in the 1–5 kDa target group. The 1–5 kDa concentrations were 93.1 wt% (140 ◦C) and 68.6 wt% (160 ◦C). This difference could be due to slightly higher retention of the group in the 30 kDa membrane at 160 ◦C (Fig. 3). When Feed-180 was processed in the mixed configuration, Fig. 4 shows that the first product Ret-10 kDa-DF was equivalent to Ret-30 kDa-DF and contained the group >30 kDa as the most abundant (61.7 wt%). The remaining molecular groups had a decreasing concentration as the molecular weight decreased. The next product was Perm-10 kDa, whose target group would be 5–10 kDa, but which was mostly in the 1–5 kDa group (51.6 wt%) partly because it was a permeate and no DF was applied. The last product was Ret-1 kDa-DF whose most abundant group was the target 1–5 kDa, with a concentration of 77.1 wt%. This value is intermediate to the obtained in Feed-140 and Feed-160, so there would be no notable differences in composition relative to obtaining this product in a cascade configuration or in a mixed configuration. 3.5. Chemical characterization and molecular weight distribution of the products Table 3 shows the main parameters characterizing each of the downstream process streams. The main parameters defining the molecular weight distribution are MW and PD. As far as these parameters are concerned, the UF/DF system has made it possible to obtain fractions with MW within the desired range given the MWCO of the membranes used. DF allowed the PD value of the fractions to be reduced, some of them to almost a commercial level, improving separation and making these fractions more defined. At 140 ◦C, starting from a highly disperse feed (18.83 kDa, PD: 31.6), four well-defined fraction were obtained: Ret-30 kDa-DF (102.75 kDa, PD: 4.0), Ret-10 kDa-DF (8.06 kDa, PD: 1.6), Ret-5 kDa-DF (6.20 kDa, PD: 1.7) and Ret-1 kDa-DF (2.59 kDa, PD: 1.2). The fractions showed low polydispersity, which is advantageous for applications where a biopolymer with defined molecular weight and narrow distribution is required. At the other temperatures, the molecular weight distributions of the feeds were also polydisperse with values of 10.98 kDa PD: 19.9 (160 ◦C) and 9.02 kDa PD: 16.2 (180 ◦C). The products had similar molecular weights to those obtained at 140 ◦C, except for the highest molecular weight retentate, which had lower MW at higher extraction temperature. At 160 ◦C, the product parameters were Ret-30 kDa-DF (59.99 kDa, PD: 3.4), Ret-10 kDa-DF (11.48 kDa, PD: 1.5), Ret-5 kDa-DF (6.85 kDa, PD: 1.4) and Ret-1 kDa-DF (4.21 kDa, PD: 1.4). At 180 ◦C the three products obtained also had adequate values: Ret-10 kDa-DF (57.06 kDa, PD: 2.4), Perm-10 kDa (5.71 kDa, PD: 1.4) and Ret-1 kDa-DF (3.33 kDa, PD: 1.5). As already mentioned, DF has the dual function of performing purification (removal of free sugars and by-products) and improving separation. This can be clearly seen in the increase of the MW value in the retentates after the application of the DF cycles. The removal of lower molecular weight hemicelluloses and pectins, and especially the removal of free sugars, resulted in this increase of the MW value accompanied by a decrease of the PD value as can be seen in Table 3. Regarding the concentration of each component, Fig. 3 and Table 3 shows that the most concentrated products are those corresponding to the highest molecular weight fraction (Ret-30 kDa-DF) where the Fig. 4. Molecular weight distribution (wt%) of the Feed and the UF/ DF products. M. Ramos-Andr´ es et al.