3.1 Characterization of extracts
Table 2 presents the composition and physical properties of extracts of Fucus vesiculosus obtained after water extraction at 120°C during 2 h. Total solids extraction yield is maximum at this temperature compared to extraction at lower temperatures between 100°C and 25°C [15]. The extracts presented a high solids content (14.31%), where fucoidans can represent 25.7% [3]. Fucose content can reach 44% of fucoidans and fucoxanthin [16], making this sugar a reliable indicator of the content of these compounds. Extracts obtained in this work presented a fucose content of 3.26% (w/w). Iodine was extracted to a content of 132 mg/L or 939 mg/kg dry solids, which is an appreciable amount that enables the enrichment of food products, taking into consideration that 150 mcg/day is the recommended daily intake for an adult [17]. Iodine is the most nutritionally valuable component of these extracts and thus the clarification of the extract and the concentration of iodine are valuable outcomes of the UF process before further use to produce pills or incorporation in other food or nutraceutical products.
The extracts are cloudy and opaque corresponding to a high turbidity of 347 NTU, which is within the range of apple processing effluent [18]. F. vesiculosus is a brown alga and thus the extract was brown with L, a* and b* of, respectively, 85.81, -0.63 and 23.75. The viscosity of the extracts was on average 1.73 mPa·s, value within the range of viscosity that of water [19]. The pH of all extracts was within the short range 5.55-5.60.
3.2 Effect of operating conditions on permeate and retentate properties
The selection of a suitable ultrafiltration membrane is essential for the optimization of the seaweed extract filtration. The performance of three membranes with cut-off of 5 and 150 kDa of polyethersulfone (PES) and of 50 kDa of hydrophilic polyethersulfone (PESH), was characterized in terms of permeate and retentate properties, and permeate flux after a volume concentration ratio (VCR) of 1.33 (Table 3). The retention coefficients are presented in Table 4.
On the retentate side, solids content increased slightly (P<0.05) (Tables 2 and 3) due to the small VCR of 1.33, with no significant differences observed between the retentates of the different membranes (Table 3). Total solids were retained between 45 to 51% by the membranes tested (Table 4).
The pH of the retentate and permeate were similar to that of the feed, as well as the color parameters.
The viscosity of retentate increased slightly but significantly (P<0.0001) compared to the feed, between 1.85 and 2.01 mPa·s. The viscosity of the permeate decreased to 1.06 mPa·s for all MWCO, close to the viscosity of water at 20°C of 1.00 mPa·s [19].
The effect on turbidity was more disparate, being 98% retained by the 5 kDa membrane and 73-79% by the other membranes. Contradictory to the change on solids content, turbidity decreased (P<0.05) on the retentate side compared to the feed when lower MWCO membranes were used. This can be explained by clogging on the membrane, which may have caused solids to be removed from the feed, but were not released into the retentate, which was not enough to affect the solids content, but was enough to affect turbidity. On the permeate side, as expected, the turbidity of the permeate decreased drastically compared to that of the feed, because these were mostly kept on the retentate and on the membrane and decreased more on the smaller MWCO membrane (P<0.05). The turbidity of the permeate prepared with 5 kDa is close to water standards, ideally <1 NTU for aesthetic aspects [20]. Higher MWCO membranes did not produce such clear streams.
The luminosity (L) increased from the feed to the permeate, the yellow tone (b*) decreased (P<0.05), and the red tone (a*) not present significant changes. The b* color parameter also decreased when lower MWCO membranes were employed (P<0.05) resulting in a weaker yellow tone. This means that high molecular weight compounds that contribute to the brown tone, stay mainly clogged in the membrane and in the retentate.
Fucose retention varied between 4 and 64%, though no significant differences were obtained.
Iodine retention was 13-16% for the PES membranes of 5 and 150 kDa and was significantly higher for the 50 kDa PESH membrane. PES membranes have been subject of improvements of hydrophilicity (called PESH membranes) to decrease the risk of the fouling caused by the high hydrophobicity of PES, especially in protein-contacting applications (Otitoju et al., 2018). However, these membranes may present disadvantages such as a high adsorption of polyphenols from olive waste, attributed to the polar interactions with the membrane [21]. These polar interactions may also explain the higher retention of iodine in this work by the PESH membrane of 50 kDa.
Overall, the performance of the membranes differs mainly from 5 kDa to 50/150 kDa, and slightly between 50 and 150 kDa, what may be due to a lack of solids with molecular weight in the range 50-150 kDa. This is in fact the case of the ethyl acetate fraction of the aqueous layer after hexane extraction of Fucus vesiculosus, in which the only 15.6% of the compounds mass presented MW between 5 and 100 kDa [22].
Based on the above-mentioned parameters, the 5 kDa PES membrane should be chosen if remotion of solids and a small retention of iodine is needed. However, other operational parameters, such as permeate flux, must be considered on the design of the industrial process of filtration.
3.3 Flux modeling
The modeling of the flux during tangential filtration is essential to optimize this process for separation efficiency, energy efficiency and equipment size. Flux modeling can be done using two approaches [23]: by applying phenomenological models such as gel-polarization, osmotic pressure, resistance-in-series, and fouling models, or using non-phenomenological models, which have been used to interpret the limiting phenomena and to predict the permeate flux.
Since the main problem in filtration processes is the membrane fouling, which results in a reduction of the permeate flux, increasing costs and decreasing productivity of the operation, fouling models were used in this work. Fouling is a result of particle-particle and particle-membrane hydrophobic and electrostatic interactions and depends on several factors such as composition of the feed (specially size of components and its aggregation behavior) and of the membrane and its physical properties, ionic strength and pH of the feed, and operation parameters such as temperature, transmembrane pressure and crossflow velocity [24, 25]. Bowen et al. (1995) suggested several steps for fouling that can occur in sequence or simultaneously:
(1) complete blocking: the small pores are blocked by particles on the membrane surface.
(2) internal pore blocking: particles adsorb to the inner surfaces of large pores.
(3) intermediate blocking: particles mount on top of others on top of the membrane with others blocking some pores.
(4) cake layer formation: a cake is built along the time as pores become blocked.
Fouling models are useful to understand what kind of mechanism affects the decrease of permeate flux and further development of strategies to diminish it. In this work, 4 fouling models were applied, each one corresponding to the 4 types of fouling previously mentioned. These models were initially developed by Hermia to dead-end filtration and further transformed to be applied to tangential filtration [14].
The models listed in Table 1 were used to describe the flux decline over time for the three ultrafiltration membranes under different conditions determined by the transmembrane pressure (TMP) and crossflow velocity (CFV) values. As an example of the fitting of all models, the flux modeling results obtained for the filtration tests with different molecular weight cut-off membranes (MWCO), at CFV=0.095 m/s and TMP=5 bar, are depicted in Fig. 1. Analyzing Fig. 1 and R2 values listed in Table 5, it can be observed that the best agreements between experimental and calculated values were achieved with the models 1, 2 and 4 with R2 in the range 0.96-0.98 for the experiments with 150 and 50 kDa membranes. The results are similar for all conditions as can be seen in the supplementary material.
Although the differences in R2 are on average 4% between models, the fittings are slightly better for model 4 (cake filtration) except for the condition 150 kDa/0.081 m·s-1 /2 bar, where model 1 (complete blocking) fitted better.
Fucus vesiculosus solids contain mainly up to 66% of carbohydrates and up to 17% of protein, depending on several environmental factors [3]. The carbohydrates include mainly alginates (up to 58.8%), then fucoidans (25.7%) and laminarians (19%). All these components are expected to be in the extract and thus to affect the filtration flux.
Filtration of alginates is a well-studied case, known to result in cake filtration [27]. The filtration flux of alginate solutions decreases rapidly, decreases faster and more if bivalent cations such Ca2+ are present due to cross-linking polymerization. Its three-dimensional structure is a factor the influences the formation of the gel layer. If its concentration is low (around 2 ppm), fouling occurs in two phases with pore blocking in the first phase followed by cake formation but on concentrations of 50 ppm, only cake formation was observed [25]. This could be the case of the filtrations in this work since a simple estimation of alginates in F. vesiculosus extracts with the contents above mentioned, can easily reach 10,000 ppm.
Fucoidans, a valuable bioactive compound present in these extracts, and the second biggest group of polysaccharides from F. vesiculosus, compared to alginate is not very viscous and it does not gel [28], thus it is not expected to contribute as much to the cake formation, though it can be trapped within the alginate cake.
An additional cause of cake filtration in several applications like whey filtration [24] or brewer’s yeast filtration [29] are proteins, which are affected by the ionic strength and also the presence of bivalent cations. Ionic strength leads to electric double layer compression which results in weak electrostatic repulsion [30]. Also, NaCl appears to reduce intermolecular reactions, promoting binary proteins deposition on the membrane surface.
Thus, it is reasonable to expect that cake filtration in the ultrafiltration of F. vesiculosus extracts is prevalent particularly due to the presence of alginates and proteins.
The predictions of the flux variation with time using the model 4, for all tested filtration conditions, are shown in Fig. 2. It is visible that the calculated flux decline profiles match quite closely with the experimental data in most cases. The parameter and R2 values for the fitting of the model 4 to the data are shown in Table 6.
A confirmation of the fouling mechanism was done by the method used by Hwang & Lin (2002). These authors plotted filtration curves of log(−dJP/dt) against log(JP) and applied a regression analysis to obtain the parameter n of the Hermia’s models (equation 2) to infer about the fouling mechanism (if close to 2, 1.5, 1 or 0 - see Table 1). An example is presented in Figure 3. The parameter n can be obtained from the linear region observable at the highest values on both axis for each curve. The second region corresponds to steady state flux and the slope tends infinite. Overall, the values obtained in this work are closer to zero (data not shown), confirming that the cake model is most suitable model to predict the ultrafiltration permeate flux of Fucus vesiculosus extracts.
3.4. Influence of MWCO, TMP and CFV
The order of magnitude of the flux and the time required for achieving its final value depends on the MWCO, TMP and CFV of the filtration process. Regarding the MWCO effect, in general it was found that permeate flux increased with membrane’s MWCO, which can be explained by the larger pore size (see Fig. 1). For instance, the initial flux for the 150 kDa membrane was of »25 L/h m2 at CFV=0.095 m/s and TMP=5 bar and was reduced around 34%. Under the same conditions, reductions in this flux of approximately 20 and 59% were obtained for the 50 and 5 kDa membranes, respectively.
The effect of TMP has influence on the driving force of the process which determines the magnitude of the permeate flux. This effect was more evident for the 5 kDa membrane (Fig. 2). When the TMP increased from 2 to 8 bar, an increase in the initial flux of » 65% and » 54% was observed operating at 0.081 m/s and 0.095 m/s, respectively. Also, the flux decline is more significant at higher TMP. The flux was lower at 8 bar when compared to one at 5 bar for the 50 kDa and 150 kDa membranes for the lowest CFV, which shows that the membranes become more fouled at higher TMP. This can probably be justified considering that at high pressures the convective transport across the membrane becomes predominant, favoring the accumulation of solutes at the membrane surface, increasing the concentration polarization and consequently fouling tendency [14]. The concentration polarization can be reduced by increasing the CFV [32]. It was found that lower fluxes were obtained when the CFV decreased, mainly for the 50 kDa and 150 kDa membranes, as expected. The change of CFV did not have a significant effect on the flux for the 5 kDa membrane. In addition, the steady state permeate flux was attained faster operating at higher CFV.