Nannochloropsis sp. Biorefinery: Recovery of Soluble Protein by Membrane Ultrafiltration/Diafiltration

This work proposes a way to maximize the potential of a Nannochloropsis sp. biorefinery process, through membrane technology, producing an extract enriched in soluble proteins, free from the insoluble protein fraction, with a low lipid content and eliminating the colored chlorophyll-a. This procedure, following the principles of a circular economy approach, allows for the valorization of a stream from the biorefining of Nannochloropsis sp. that, otherwise, would be considered a residue without commercial value. The process proposed minimizes fouling phenomena at the membrane surface, making it possible to achieve high permeate fluxes, thus reducing the need for membrane cleaning and, therefore, contributing to an extended membrane lifetime. Supernatant obtained after centrifugation of a suspension of ruptured Nannochloropsis sp. cells was processed by ultrafiltration using a membrane with a cut-off of 100 kDa MWCO. Two different operating approaches were evaluated—controlled transmembrane pressure and controlled permeate flux—under concentration and diafiltration modes. Ultrafiltration operated in a diafiltration mode, under controlled permeate flux conditions, led to the highest soluble protein recovery (78%) with the highest constant permeate flux (12 L·m−2·h−1) and low membrane fouling.


Introduction
Proteins are among the microalgae products that have been receiving great attention from the research community. Microalgae proteins are a potential suitable source of highquality protein not only for human food [1] but also for animal feed [2], and in specific cases, such as the case of phycobiliproteins, for nutraceutical and pharmaceutical applications [3].
Several species of microalgae are known for their high protein contents, with nutritional values similar to other traditional protein sources, such as meat, egg, soybean, and milk [4,5]. Fradique et al. [1] reported having successfully produced a high-protein pasta product through the incorporation of Chlorella vulgaris and Spirulina maxima in semolina flour. The authors also reported that the cooking quality of pasta was not compromised. Nevertheless, the incorporation of microalgae protein in food products faces several challenges, mostly those related to their organoleptic characteristics, such as odor, taste, and color [6]. Color is the first parameter noticed by the consumer and the common green color, linked to the presence of chlorophyll, leads to very low sensorial acceptance by consumers [7]. Qazi et al. [8] studied the impact on dough rheology and bread quality after and 1000 kDa) at a fixed TMP (2.07 bar). The best performance was achieved with the 300-kDa membrane, and a protein recovery of 37% (w/w total proteins in the supernatant) was reported [26]. This low protein recovery could be explained by fouling phenomena, which reduce the permeate flow rate and increase the retention by the membrane.
The severe fouling observed at the start of the constant transmembrane pressure (TMP) ultrafiltration process, due to the often very high initial flux of the clean membrane, may be reduced by applying a milder operation, using a controlled permeate flux strategy, operated under subcritical/sustainable permeate flux conditions [27,28]. Monte et al. studied the harvesting of carotenoid-rich Dunaliella salina through ultrafiltration in a concentration mode under controlled flux conditions [29]. Similarly, Serra et al. combined ultrafiltration in a diafiltration mode with controlled flux conditions for the purification of viscous arabinoxylans from corn fiber through dia-ultrafiltration [30].
In the present work, membrane fouling occurring during protein filtration could be reduced by operating the ultrafiltration process in a diafiltration mode. The dia-ultrafiltration process enhances the removal of small proteins and carbohydrates, preventing the viscosity of the feed solution from rising over time.
The main objective of this study was to define a suitable methodology for using membrane processing to obtain a high permeate flux with maximum soluble protein recovery (free from chlorophyll-a, insoluble proteins, and lipids). This work also aimed to establish a process with minimum fouling, which could lead to easy membrane cleaning, thereby also contributing to maximizing the membrane lifetime. The global efficiency of the process under study was mainly quantified by the permeate volumetric flux, J v (L·m −2 ·h −1 ), and by the percentage of soluble protein recovered in the permeate. To our knowledge, ultrafiltration in a diafiltration operation mode under controlled permeate flux has not been reported and evaluated as a methodology for the recovery of soluble proteins in microalgae biorefineries. Besides the study performed by dia-ultrafiltration, ultrafiltration in a concentration operation mode was also tested. In both cases, two different operating conditions were compared, corresponding to controlled transmembrane pressure and controlled permeate flux.

Materials
The Nannochloropsis sp. biomass used in this work was cultivated in photobioreactors (PBRs), mechanically disrupted and supplied by A4F-Algae for Future (Portugal). For the ultrafiltration studies, the Nannochloropsis sp. disrupted biomass (100 g·L −1 , on a salt-free dry weight basis-SFDW) was centrifuged at 11,000× g (for 2 × 20 min, 5 • C) and the collected supernatant (30 g SFDW ·L −1 ) was analyzed in terms of its salt-free dry weight, proteins, lipids, and chlorophyll-a content.
Previous studies of ultrafiltration using a hydrophilic ceramic membrane with a nominal molecular weight cut-off (MWCO) of 300 kDa revealed a significant volumetric flux decrease during the process, which may be related to high fouling, namely intrapore fouling (data not shown here). Nannochloropsis sp. contains various proteins, which are associated with light-harvesting complexes, with molecular masses ranging between 21 and 32 kDa [31]. Therefore, for optimization of soluble protein recovery, a hollow fiber module containing a tighter ultrafiltration membrane with an MWCO of 100 kDa was used (GE Healthcare, Chicago, IL, USA; model: UFP-100-C-5A). It was composed of 520 hydrophilic polysulfone (PS) fibers, with an internal diameter of 0.5 mm, length of 0.32 m, and total filtration area of 0.2 m 2 . The membrane selected is hydrophilic, to enhance the permeation of soluble proteins, as the extract is aqueous, and to avoid protein denaturation that is prone to occurring at the surface of hydrophobic materials. The membrane-cleaning products Ultrasil 110 and Ultrasil 75 were purchased from Ecolab (Mississauga, Ontario, Canada) and ethanol (97%) was obtained from Panreac (Barcelona, Spain). For analytical procedures, acetone 90.0% and methanol 99.8% were purchased from JMGS (Odivelas, Portugal), chloroform was obtained from Honeywell/Riedel-de Haën (Seelze, Germany), and the Pierce™ BCA Protein Assay Kit was acquired from Thermo Fisher Scientific (Waltham, MA, USA).

Membrane Processing
In this work, the supernatant (30 g SFDW ·L −1 ) from the centrifugation of Nannochloropsis sp. disrupted biomass and its dilution, with a dilution of 1:3 corresponding to 10 g SFDW ·L −1 , were processed with the selected membrane system, as summarized in Table 1. The supernatant (30 g SFDW ·L −1 and 60 g DW ·L −1 ) from the centrifugation of Nannochloropsis sp. disrupted biomass was diluted with Nannochloropsis sp. culture medium, which has a salt concentration of 30 g DW ·L −1 , to a dilution ratio of 1:3. After the dilution, the concentration of the diluted supernatant was 10 g SFDW ·L −1 and 40 g DW ·L −1 . Primarily, the membrane ultrafiltration experiments were conducted in a concentration operation mode, under controlled transmembrane pressure (TMP) conditions, as it is the most commonly reported in the literature (see the scheme of membrane setup in Figure 1a). The membrane filtration unit constituted a feed tank (1), recirculation pump (2), pressure gauges (3), membrane module (4), and permeate tank (5). Additionally, the ultrafiltration experiments were also performed in a diafiltration operation mode followed by concentration, under controlled transmembrane pressure conditions (see the scheme of membrane set-up in Figure 1b), where a water tank (6) and a peristaltic pump, for a water feed, were added. The value of transmembrane pressure was set as 0.2 bar (see Table 1) to preserve mild operating conditions. permeation of soluble proteins, as the extract is aqueous, and to avoid protein denaturation that is prone to occurring at the surface of hydrophobic materials. The membranecleaning products Ultrasil 110 and Ultrasil 75 were purchased from Ecolab (Mississauga, Ontario, Canada) and ethanol (97%) was obtained from Panreac (Barcelona, Spain). For analytical procedures, acetone 90.0% and methanol 99.8% were purchased from JMGS (Odivelas, Portugal), chloroform was obtained from Honeywell/Riedel-de Haën (Seelze, Germany), and the Pierce™ BCA Protein Assay Kit was acquired from Thermo Fisher Scientific (Waltham, MA, USA).

Membrane Processing
In this work, the supernatant (30 gSFDW·L −1 ) from the centrifugation of Nannochloropsis sp. disrupted biomass and its dilution, with a dilution of 1:3 corresponding to 10 gSFDW·L −1 , were processed with the selected membrane system, as summarized in Table 1. The supernatant (30 gSFDW·L −1 and 60 gDW·L −1 ) from the centrifugation of Nannochloropsis sp. disrupted biomass was diluted with Nannochloropsis sp. culture medium, which has a salt concentration of 30 gDW·L −1 , to a dilution ratio of 1:3. After the dilution, the concentration of the diluted supernatant was 10 gSFDW·L −1 and 40 gDW·L −1 . Primarily, the membrane ultrafiltration experiments were conducted in a concentration operation mode, under controlled transmembrane pressure (TMP) conditions, as it is the most commonly reported in the literature (see the scheme of membrane setup in Figure 1a). The membrane filtration unit constituted a feed tank (1), recirculation pump (2), pressure gauges (3), membrane module (4), and permeate tank (5). Additionally, the ultrafiltration experiments were also performed in a diafiltration operation mode followed by concentration, under controlled transmembrane pressure conditions (see the scheme of membrane set-up in Figure 1b), where a water tank (6) and a peristaltic pump, for a water feed, were added. The value of transmembrane pressure was set as 0.2 bar (see Table 1) to preserve mild operating conditions. As shown in Table 1, three experiments were also performed under controlled permeate flux conditions, where the permeate flux was fixed and the TMP was allowed to vary. The membrane setup for the controlled permeate flux experiments was similar to the preceding one (setup for the controlled TMP experiments), both for concentration and diafiltration operation modes (see the scheme of membrane setup in Figure 1c,d, respectively). When operating under controlled permeate flux, its value was imposed by introducing a positive displacement pump (5) in the permeate circuit, which controls and defines the permeate flux, as presented in Figure 1c,d. The value of permeate flux was set at 12 L·m −2 ·h −1 (see Table 1). This value was selected after preliminary experiments, which showed that at this flux, no increase of transmembrane pressure occurs (sub-critical, sustainable flux conditions).
In each membrane experiment, the feed was pumped from the feed tank through the membrane module using a peristaltic pump and recirculated to the feed tank, with a cross-flow velocity of 0.25 m·s −1 . The feed reservoir was filled with 1.5 ± 0.5 L of protein aqueous extract at 20 • C and pH 6.7 ± 0.1. The permeate was collected in the permeate tank and the permeate flux was monitored by mass acquisition using an electronic balance (Kern 572, Kern, Balingen, Germany). For the ultrafiltration in a diafiltration operation mode, the diafiltration volume D (-) was calculated through Equation (1): where V water added (L) is the volume of the water (diafiltration solvent) added and V feed (L) is the volume of the feed. For the ultrafiltration process operated in a concentration operation mode, the concentration factor, CF (-), was determined according to Equation (2): where m t0 (kg) represents the mass in the feed compartment in the beginning of the experiment and m permeate (kg) is the permeate mass collected during a given period. The permeate volumetric flux, permeance, apparent rejection of soluble proteins, and percentage of soluble protein recovered in the permeate were calculated during the membrane ultrafiltration experiments.
The volumetric permeate flux J v (L·m −2 ·h −1 ) was calculated according to Equation (3): where m permeate (kg) is the mass of permeate, ρ (kg·m −3 ) is the density of the permeate, A (m 2 ) is the total membrane filtration area, and t (h) is the time of permeation. The permeance L p (L·m −2 ·h −1 ·bar −1 ) was calculated through Equation (4): where J v (L·m −2 ·h −1 ) is the volumetric permeate flux and TMP (bar) is the transmembrane pressure. In each test (controlled transmembrane pressure concentration/diafiltration and controlled permeate flux concentration/diafiltration), the global apparent rejection of a target compound R i (%) and the percentage of soluble protein recovered in the permeate were calculated during the filtration experiments through Equations (5) and (6), respectively: % Recovery s.protein = m protein, perm (t) m protien, f eed (t 0 ) where in Equation (5), C i,perm (g·L −1 ) is the concentration of the target compound in the collected permeate and C i,feed (g·L −1 ) is the concentration of the same compound in the feed (retentate) compartment, and in Equation (6), m s.protein,perm (t) (g) is the mass of soluble protein in the permeate compartment at a defined time of the experiment (t) and m s.protein,feed (t 0 ) (g) is the mass of soluble protein in the feed compartment at the initial time of the experiment. The soluble protein loss was defined as the soluble protein that is not recovered in the permeate for each experiment and was evaluated through a soluble protein mass balance, as shown in Equation (7): where m s.protein,feed (t 0 ) (g) is the mass of soluble protein in the feed compartment at the initial time of the experiment, m s.protein,perm (t F ) (g) is the mass of soluble protein recovered in the permeate at the end of the experiment, m s.protein,reten (t F ) (g) is the mass of soluble protein in the retentate at the end of the experiment, and m s.protein,accum (t F ) (g) is the soluble protein adsorbed/accumulated on the membrane surface and/or pores (surface intra-pore fouling).
The fouling phenomena, defined as the undesirable adsorption/deposition of dissolved solutes or suspended particles, was evaluated through the resistance-in-series model [32]: where in Equation (8), J v (m·s −1 ) is the volumetric permeate flux, TMP (Pa) the transmembrane pressure, η (Pa·s) the dynamic viscosity of the permeate (we considered the water viscosity at 20 • C, 1.002 × 10 −3 Pa·s, since we are dealing with an aqueous extract), and R total (m −1 ) represents the total resistance. We considered that the total resistance R total (m −1 ) results from a series of resistances introduced by the intrinsic membrane resistance R m (m −1 ) and the resistance caused by the fouling, which may comprise two distinct contributions: R rev (m −1 ), standing for the resistance that disappears when the pressure across the membrane is released and pure water is passed through the system at the end of the experiment, removing unbounded solutes from the membrane surfacei.e., reversible fouling resistance; R irrev (m −1 ), the resistance that is only removed when a cleaning cycle using chemical agents is applied, removing the compounds chemically bonded to the membrane-i.e., irreversible fouling resistance. Therefore, we have then for the total resistance: The resistance that the membrane offers to the permeation of pure water, R m (m −1 ), was calculated through the hydraulic permeance measured at the beginning of each experiment: where Lp w (m·Pa −1 ·s −1 ) is the hydraulic permeance and η w (Pa·s) is the viscosity of water at 20 • C (1.002 × 10 −3 Pa·s).
To estimate the remaining resistances, as mentioned above (9), the total resistance, R total (m −1 ) was calculated at the end of each Nannochloropsis sp. supernatant filtration experiment, using the last measured value of volumetric permeate flux, J v (m·s −1 ), and R m + R irrev (m −1 ) values were obtained through the volumetric flux of water, measured after a flush with pure water (20 • C) to eliminate reversible fouling. By subtracting R m + R irrev (m −1 ) from R total (m −1 ), it was then possible to calculate R rev (m −1 ) and compute R irrev (m −1 ), subtracting the known parameters, R m (m −1 ) and R rev (m −1 ), from the total resistance R total (m −1 ).

Membrane Cleaning
To achieve reproducible membrane experiments, after membrane cleaning, the hydraulic permeance was measured. The hydraulic permeance measured with deionized water at 20 • C before the experiments was 221 ± 3 L/(m 2 h bar). After each trial, the membrane was cleaned, using a pre-established cleaning cycle to recover the initial hydraulic permeance. The cleaning procedure comprised the following steps: use of an alkaline solution (at 45 ± 2 • C) with a concentration of 0.5% (w/w) of Ultrasil 110 recirculated for 30 min, to remove organic fouling; then, an acid solution of Ultrasil 75 with a concentration of 0.03% (w/w) was recirculated for 30 min (at 45 ± 2 • C) to remove the inorganic fouling; the last step consisted of circulating ethanol (at 45 ± 2 • C) for 20 min with a concentration of 70% (v/v), to recover the hydrophilicity of the membrane surface and restore the water permeability [33]. Between these different cleaning operations and steps, the membrane was flushed with deionized water until the pH was stabilized at 7 and negligible conductivity was reached.

Dry Weight and Salinity Quantification
The dry weight of each sample was determined by the weight difference before and after water evaporation. The drying program was run at 105 • C until a constant weight was achieved in a moisture analyzer (Kern DBS 60-3, Kern, Germany). To determine the sample salt-free dry weight (SFDW), a salinity refractometer (Kern ORA 1SA, Kern, Germany) was used.

Protein Quantification
Total protein quantification of the different samples was performed using a Pierce™ BCA Protein Assay Kit (Thermo Scientific, Waltham, MA, USA). The calibration curves and quantification assays were performed according to the supplier's instructions. Soluble protein quantification was performed after centrifugation (14,000× g for 15 min) of each sample to precipitate the membrane proteins that remained attached to cellular debris. The proteins that remained in the supernatant after centrifugation were considered soluble proteins and analyzed through a bicinchoninic acid (BCA) assay [34,35].

Total Lipids Quantification
The total lipid content was extracted with a solvent mixture of 1:1 methanol:chloroform, as described by Ryckebosch et al. [36]. Briefly, 100 mg of previously freeze-dried samples was mixed with 4 mL of methanol. Chloroform (2 mL) and water (0.4 mL) were then added and the sample was vortex-mixed for 45 s. Chloroform (2 mL) and water (2 mL) were added again and the sample was vortex-mixed for another 45 s. After that, the sample was centrifuged (4000× g for 10 min). The upper layer was reserved in a clear tube and the remaining pellet was re-extracted with 4 mL of methanol:chloroform 1:1. The centrifugation step was repeated and the solvent layers were passed through a layer of anhydrous sodium sulfate. Afterward, the solvent was removed by rotary evaporation and the lipid content was determined gravimetrically.

Chlorophyll-a Quantification
Quantification of chlorophyll-a was performed through spectrophotometric analysis. Briefly, 1 mg of a freeze-dried sample was mixed with 4 mL 90.0% acetone and vortex-mixed for 60 s. Then, the sample was centrifuged at 4000× g (for 10 min, 5 • C). The extracted chlorophyll-a, recovered in the supernatant, was analyzed by spectrophotometry (Genesys 10S UV-VIS), and the chlorophyll-a content was determined according to Equation (11), proposed by Ritchie et al. [37], for two different wavelengths: Membranes 2022, 12, 401 8 of 17

Results and Discussion
The analysis of the results obtained is focused here on the ultrafiltration process, after centrifugation of disrupted Nannochloropsis sp. biomass. In Section 3.1, characterization of the supernatant from the centrifugation of disrupted Nannochloropsis sp. biomass is presented. Subsequently, in Section 3.2, results from the experimental work performed with ultrafiltration/diafiltration are presented and discussed.

Nannochloropsis sp. Supernatant Characterization
The disrupted (>90% disintegration) Nannochloropsis sp. biomass (100 g SFDW ·L −1 ) was centrifuged and the supernatant was collected and analyzed in terms of the salt-free dry weight, proteins, lipids, and chlorophyll-a. The supernatant samples, used as feed solutions for filtration experiments without dilution, presented on average a salt-free dry weight of 30.2 ± 0.8 g·L −1 . The supernatant samples' characterization ( Table 2) revealed that 83% of the proteins (w/w total proteins) present in the supernatant were soluble. The total lipids analysis shows that the supernatant contained 4.37 ± 0.22 g·L −1 lipids, on average. The Nannochloropsis sp. used in the present study was grown under N-replete conditions; lipids were mainly present in the form of polar lipids (glycolipids and phospholipids), which are crucial to preserving cell-membrane integrity [20]. These lipid classes are expected to have interactions with cytoskeletal proteins [21] that are responsible for the structural organization of the cell and to ensure several cellular processes (e.g., cell division, motility, and contractility) [38]. Therefore, some of the polar lipids attached to cytoskeletal proteins might remain in the aqueous supernatant.

Ultrafiltration/Diafiltration
During the ultrafiltration experiments, different operation modes (concentration and diafiltration) and operating conditions were compared (controlled transmembrane pressure, TMP, and controlled permeate flux). In an ultrafiltration process under controlled TMP, the transmembrane pressure is fixed and the flux is allowed to evolve, declining as the membrane fouls. On the other hand, ultrafiltration under controlled permeate flux allows for the precise control of convective transport toward the membrane surface, making it possible to adjust the viscous force near the membrane [28,39,40]. If the controlled permeate flux is imposed assuring a sustainable flux (subcritical conditions), membrane fouling is avoided, or at least, minimized. In fact, unlike the controlled permeate flux operation, the controlled transmembrane pressure operation is more prone to causing severe fouling phenomena, with a typical trend for high initial permeate flux, followed by a drastic reduction [27].
The performance of the ultrafiltration/diafiltration process for both operating conditions is presented below.

Volumetric Flux and Permeance
The results for the volumetric flux (J v ) obtained, while maintaining a controlled transmembrane pressure at 0.2 bar or a controlled permeate flux of around 12 L·m −2 ·h −1 , are presented in Figure 2. The volumetric flux decreases rapidly at the beginning of the process for the ultrafiltration under controlled TMP, from over 8 to 4 L·m −2 ·h −1 (the permeate flux decline is extremely fast in the beginning, and after the 5 min required to measure the permeate flux, this value is already significantly lower than its initial clean water flux). This behavior is a consequence of the intensive fouling that occurs at the start of a constant TMP ultrafiltration operation [27,39]. These results are comparable with those obtained by Safi et al. [26], where the permeate flow rate decreased at the beginning of the ultrafiltration process, as measured using a 300 kDa MWCO membrane under a constant TMP of 2.07 bar. The ultrafiltration process under controlled permeate flux ( Figure 2B) is known to provide milder operating conditions when compared to the ultrafiltration process under controlled TMP (Figure 2A), allowing a higher value of volumetric flux, at around 12 L·m −2 ·h −1 , to be maintained during the entire process. The value of 12 L·m −2 ·h −1 , selected for the imposed permeate flux, was based on previous experience in the processing of biological media (data not shown here).
presented in Figure 2. The volumetric flux decreases rapidly at the beginning of the process for the ultrafiltration under controlled TMP, from over 8 to 4 L·m −2 ·h −1 (the permeate flux decline is extremely fast in the beginning, and after the 5 min required to measure the permeate flux, this value is already significantly lower than its initial clean water flux). This behavior is a consequence of the intensive fouling that occurs at the start of a constant TMP ultrafiltration operation [27,39]. These results are comparable with those obtained by Safi et al. [26], where the permeate flow rate decreased at the beginning of the ultrafiltration process, as measured using a 300 kDa MWCO membrane under a constant TMP of 2.07 bar. The ultrafiltration process under controlled permeate flux ( Figure 2B) is known to provide milder operating conditions when compared to the ultrafiltration process under controlled TMP (Figure 2A), allowing a higher value of volumetric flux, at around 12 L·m −2 ·h −1 , to be maintained during the entire process. The value of 12 L·m −2 ·h −1 , selected for the imposed permeate flux, was based on previous experience in the processing of biological media (data not shown here). Similar to the volumetric flux (Figure 2A), the membrane calculated permeance, considering the osmotic pressure difference as negligible, followed the same trends for the controlled TMP experiments ( Figure 3A). The selection of an ultrafiltration process under controlled permeate flux provides better results in terms of the process operating performance. Still, it should be noted that even with a gentle permeate flux of 12 L·m −2 ·h −1 (experiments 4 and 5), the transmembrane pressure increased due to some fouling, and consequently, the permeance decreased. Similar to the volumetric flux (Figure 2A), the membrane calculated permeance, considering the osmotic pressure difference as negligible, followed the same trends for the controlled TMP experiments ( Figure 3A). The selection of an ultrafiltration process under controlled permeate flux provides better results in terms of the process operating performance. Still, it should be noted that even with a gentle permeate flux of 12 L·m −2 ·h −1 (experiments 4 and 5), the transmembrane pressure increased due to some fouling, and consequently, the permeance decreased.

Soluble Protein Recovery and Permeate Characterization
The effects of different operation modes (concentration/diafiltration) and operating conditions (controlled TMP and controlled permeate flux) on the recovery of soluble protein in the permeate can be observed in Table 3, as calculated by Equation (6) (see Section 2.2). The experiments under controlled permeate flux exhibit higher values of protein recovery in the permeate (between 44% and 62% for concentration mode and 78% for diafiltration mode) than the experiments conducted under controlled TMP (between 35% and 40% for concentration mode and 58% for diafiltration mode). The lower protein recovery obtained for controlled TMP experiments is related to the more severe fouling associated with these process operating conditions, as it is known that fouling can act in the selectivity/rejection of solutes during the filtration process [27,39].
x FOR PEER REVIEW 10 of 17

Soluble Protein Recovery and Permeate Characterization
The effects of different operation modes (concentration/diafiltration) and operating conditions (controlled TMP and controlled permeate flux) on the recovery of soluble protein in the permeate can be observed in Table 3, as calculated by Equation (6) (see Section 2.2). The experiments under controlled permeate flux exhibit higher values of protein recovery in the permeate (between 44% and 62% for concentration mode and 78% for diafiltration mode) than the experiments conducted under controlled TMP (between 35% and 40% for concentration mode and 58% for diafiltration mode). The lower protein recovery obtained for controlled TMP experiments is related to the more severe fouling associated with these process operating conditions, as it is known that fouling can act in the selectivity/rejection of solutes during the filtration process [27,39].  The results obtained in terms of protein recovery during the diafiltration under controlled TMP are slightly lower than those obtained by Balti et al. [41] for the diafiltration of soluble proteins from Spirulina. The authors reported 64% recovery of soluble proteins in the permeate, using a 150 kDa ceramic membrane at a fixed TMP of 4 bar. It is also important to note that the authors applied two more diavolumes (a total of five) than those applied during the diafiltration described in the present work (a total of three), which might have contributed to a slightly higher protein recovery.
The mass balance for soluble proteins (Table 4) also reinforces the association of lower soluble protein recovery with more severe fouling, due to soluble protein adsorption. The experiments under controlled TMP exhibit higher soluble protein losses due to adsorption (OUT Adsorbed /IN Feed ) to the membrane during the filtration process (between 42 ± 0.4% and 35 ± 1.5% for concentration mode and 37 ± 2.0% for diafiltration mode), than the experiments under controlled permeate flux (between 37 ± 2.7% and 13 ± 2.6% for concentration mode and 7 ± 0.4% for diafiltration mode). These results also confirm that it is easier to clean a membrane after operation under controlled permeate flux than under controlled TMP.  The results presented in Table 3 show that the diafiltration experiments lead to higher soluble protein recovery in the permeate. A comparison of the two diafiltration experiments (Exp.3-under controlled TMP and Exp.6-under controlled permeate flux) is shown in Figure 4. Comparison of these two experiments reveals that the soluble protein recovery is higher among all diafiltration volumes when the experiment is operated under controlled permeate flux ( Figure 4A). Figure 4B shows that the diafiltration + concentration under controlled permeate flux presents clear advantages over the diafiltration + concentration under controlled TMP, not only in terms of soluble protein recovery in the permeate (58% for Exp.3-under controlled TMP and 78% for Exp.6-under controlled permeate flux) but also in the reduction of the overall process time.
The characterization of the permeates after each trial shows that, for all the experiments, insoluble proteins, lipids, and chlorophyll-a were retained in the retentate (Table 5), which is comparable with the results obtained in other ultrafiltration studies [25,26]. Polysaccharides may have been retained, according to other studies [26,42], as expected, given their large molecular weights. Chlorophyll-a retention may be explained by its adsorption to small hydrophobic cell debris that remained in the supernatant but was larger than the membrane pores. The results obtained in terms of chlorophyll-a retention are comparable with those reported by Safi et al. [26] when using a 300 kDa PES membrane, operated under a controlled TMP. The authors reported that the chlorophyll present in a Nannochloropsis gaditana supernatant was fully retained and associated the result with the fact that chlorophyll was present in lipid droplets that were larger than the membrane cut-off. The results presented in Table 5 also show that the lipids present in the supernatant were highly retained, which might have contributed to the retention of chlorophyll-a. operated under a controlled TMP. The authors reported that the chlorophyll present in a Nannochloropsis gaditana supernatant was fully retained and associated the result with the fact that chlorophyll was present in lipid droplets that were larger than the membrane cutoff. The results presented in Table 5 also show that the lipids present in the supernatant were highly retained, which might have contributed to the retention of chlorophyll-a.   (2) ND (2) (1) Detection limit: 5 mg·L −1 ; (2) Detection limit: 3 µg·L −1 .
This is an interesting result since the green color is a major drawback for certain food  operated under a controlled TMP. The authors reported that the chlorophyll present in a Nannochloropsis gaditana supernatant was fully retained and associated the result with the fact that chlorophyll was present in lipid droplets that were larger than the membrane cutoff. The results presented in Table 5 also show that the lipids present in the supernatant were highly retained, which might have contributed to the retention of chlorophyll-a.  1.53 ± 0.13 0.20 ± 0.05 0.18 ± 0.05 Chlorophyll-a (g·L −1 ) 0.04 ± 0.01 ND (2) ND (2) (1) Detection limit: 5 mg·L −1 ; (2) Detection limit: 3 µg·L −1 .
This is an interesting result since the green color is a major drawback for certain food applications. These results are important not only for the soluble protein fraction enrichment but also for the overall performance of a Nannochloropsis biorefinery. The retention of more than 87% of the lipids present in the supernatant, in all the assays, is extremely significant for increasing the overall biorefinery efficiency as the membrane retentate can  operated under a controlled TMP. The authors reported that the chlorophyll present in a Nannochloropsis gaditana supernatant was fully retained and associated the result with the fact that chlorophyll was present in lipid droplets that were larger than the membrane cutoff. The results presented in Table 5 also show that the lipids present in the supernatant were highly retained, which might have contributed to the retention of chlorophyll-a.  1.53 ± 0.13 0.20 ± 0.05 0.18 ± 0.05 Chlorophyll-a (g·L −1 ) 0.04 ± 0.01 ND (2) ND (2) (1) Detection limit: 5 mg·L −1 ; (2) Detection limit: 3 µg·L −1 .
This is an interesting result since the green color is a major drawback for certain food applications. These results are important not only for the soluble protein fraction enrichment but also for the overall performance of a Nannochloropsis biorefinery. The retention of more than 87% of the lipids present in the supernatant, in all the assays, is extremely significant for increasing the overall biorefinery efficiency as the membrane retentate can  0.08 ± 0.02 ND (2) ND (2) ND (2) ND (2)  0.04 ± 0.01 ND (2) ND (2) (1) Detection limit: 5 mg·L −1 ; (2) Detection limit: 3 µg·L −1 .
This is an interesting result since the green color is a major drawback for certain food applications. These results are important not only for the soluble protein fraction enrichment but also for the overall performance of a Nannochloropsis biorefinery. The retention of more than 87% of the lipids present in the supernatant, in all the assays, is extremely significant for increasing the overall biorefinery efficiency as the membrane retentate can incorporate the feed stream for a lipid extraction process. The lipids' retention can be related to the retention of insoluble compounds, such as insoluble proteins, but also to the formation of aggregates between lipids and structural proteins with a mass higher than the membrane molecular weight cut-off [21,42].

Fouling and Membrane Cleaning
Membrane fouling during membrane operation is inevitable, and therefore, the easiness and readiness of membrane cleaning is an important feature that needs to be taken into account for industrial membrane operation. The soluble protein mass balance (presented in Section 3.2.2), hydraulic permeance decrease, and resistance (reversible and irreversible) for the different experiments were analyzed to evaluate the possible relationships between the different operating conditions and the easiness/readiness of membrane cleaning. The initial hydraulic permeance Lp w,i (L/(h·m 2 ·bar)), experimentally determined before each of the ultrafiltration experiments, and the final hydraulic permeance Lp w,f (L/(h·m 2 ·bar)), determined at the end of each experiment, after a hydraulic flush to remove the layer deposited on the membrane surface (reversible fouling), were evaluated and the results are shown in Table 6. The experiments under controlled permeate flux exhibit slightly lower values of hydraulic permeance loss, % Lp w Loss (1 − Lp w,f /Lp w,i ), than the experiments under controlled TMP. The last permeance values of each experiment with Nannochloropsis sp. supernatant Lp f (L/(h·m 2 ·bar)) are shown in Table 6, confirming that experiments under controlled permeate flux can maintain a higher permeance value until the end of the experiment. On the other hand, comparing the diafiltration operation mode with the concentration operation mode (Table 6), it is notable that after a diafiltration step, the permeance value remains higher than after a concentration step. The difference between the two operation modes becomes even clearer when the operations are performed under controlled permeate flux. These data indicate that fouling phenomena are more severe in ultrafiltration operated at a concentration mode under controlled TMP, and as a consequence, membrane cleaning might be harder under these conditions. Likewise, the membrane lifetime may be reduced when the ultrafiltration experiments are performed under controlled TMP. Severe fouling during ultrafiltration would occur if the operating conditions selected favored a strong convective transport of foulant toward the membrane surface and the build-up of high local concentrations of these foulants near the membrane surface. In the particular case studied in this work, the presence of biological material, namely proteins and other biological compounds, represents a severe risk of fouling unless adequate operating strategies are adopted (operation under controlled permeate flux). Therefore, diafiltration followed by a concentration step under controlled permeate flux seems to be the best combination of operation conditions, allowing for lower membrane fouling and easier membrane cleaning.
Furthermore, the results obtained for the calculated resistance due to reversible R rev (m −1 ) and irreversible fouling R irrev (m −1 ) for each experiment are shown in Figure 5. Reversible resistance is slightly higher in the experiments under controlled TMP than in the experiments under controlled permeate flux, which is in accordance with other fouling studies [27,43]. This result suggests that it might be easier to clean a membrane, by hydraulic flushing, after ultrafiltration under controlled permeate flux than under controlled TMP. Such a result can be explained by the differences in the deposited layer's thickness. Vyas et al. [43] reported that the deposited layer formed on the membrane surface during constant permeate flux experiments was very thin. The experiments under controlled TMP exhibit higher values for the irreversible resistance, which might implicate the increase in frequency (and/or intensity) of chemical cleaning cycles, and consequently, the time devoted to cleaning protocols. The membrane lifetime may also be affected by the higher irreversible fouling experienced in the ultrafiltration process under controlled TMP.
Membranes 2022, 12, x FOR PEER REVIEW 14 of 17 particular case studied in this work, the presence of biological material, namely proteins and other biological compounds, represents a severe risk of fouling unless adequate operating strategies are adopted (operation under controlled permeate flux). Therefore, diafiltration followed by a concentration step under controlled permeate flux seems to be the best combination of operation conditions, allowing for lower membrane fouling and easier membrane cleaning. Furthermore, the results obtained for the calculated resistance due to reversible Rrev (m −1 ) and irreversible fouling Rirrev (m −1 ) for each experiment are shown in Figure 5. Reversible resistance is slightly higher in the experiments under controlled TMP than in the experiments under controlled permeate flux, which is in accordance with other fouling studies [27,43]. This result suggests that it might be easier to clean a membrane, by hydraulic flushing, after ultrafiltration under controlled permeate flux than under controlled TMP. Such a result can be explained by the differences in the deposited layer's thickness. Vyas et al. [43] reported that the deposited layer formed on the membrane surface during constant permeate flux experiments was very thin. The experiments under controlled TMP exhibit higher values for the irreversible resistance, which might implicate the increase in frequency (and/or intensity) of chemical cleaning cycles, and consequently, the time devoted to cleaning protocols. The membrane lifetime may also be affected by the higher irreversible fouling experienced in the ultrafiltration process under controlled TMP.

Conclusions
The ultrafiltration process studied, using a 100 kDa MWCO membrane, resulted in obtaining an enriched protein fraction, free from potentially undesirable contaminants. The results show that the ultrafiltration process operated under controlled permeate flux presents significant advantages over operation under controlled transmembrane pressure: higher permeate flux (12 L·m −2 ·h −1 ), higher soluble protein recovery (78% in a diafiltration mode), and lower membrane fouling. This work shows that when working under controlled permeate flux, the operator determines the convective transport conditions of the feed stream toward the surface of the membrane. This means that the operator allows a defined mass flux of feed across the membrane, avoiding the situation that occurs during controlled transmembrane pressure operation where a very high-mass flux of feed might occur during the initial period of operation (when the membrane is still clean and not

Conclusions
The ultrafiltration process studied, using a 100 kDa MWCO membrane, resulted in obtaining an enriched protein fraction, free from potentially undesirable contaminants. The results show that the ultrafiltration process operated under controlled permeate flux presents significant advantages over operation under controlled transmembrane pressure: higher permeate flux (12 L·m −2 ·h −1 ), higher soluble protein recovery (78% in a diafiltration mode), and lower membrane fouling. This work shows that when working under controlled permeate flux, the operator determines the convective transport conditions of the feed stream toward the surface of the membrane. This means that the operator allows a defined mass flux of feed across the membrane, avoiding the situation that occurs during controlled transmembrane pressure operation where a very high-mass flux of feed might occur during the initial period of operation (when the membrane is still clean and not fouled). As a consequence, a wise choice of permeate flux, imposed and controlled by the operator, allows for working under subcritical (or sustainable) permeate flux conditions, minimizing fouling and thus extending the operation under manageable permeate flux values. The combination of ultrafiltration in a diafiltration mode followed by concentration under controlled permeate flux conditions enhances the soluble protein recovery in the permeate and allows a higher value of volumetric flux to be maintained for the entire diafiltration process. This strategy led to an increase in the average flux (3×) and soluble protein recovery (+34%), along with a decrease in soluble protein rejection (to half of the initial value) and in soluble protein adsorbed to the membrane (to 17% of the initial value). This strategy may be applied to other microalgae in a very straightforward way. It may also be applicable to membrane processes that suffer from significant membrane fouling, e.g., when the feed solution has lipids and other potential foulants.