Impact of Diafiltration Media and Filtration Modes on Fouling and Performance in Skim Milk Microfiltration: A Comparative Study

This research utilized a customized laboratory setup to compare the filtration performance and fouling buildup during microfiltration with polymeric membranes of skim milk using 2 diafiltration media: ultrafiltration permeate and ultrapure water. Two filtration modes were evaluated: in stage 1, the diafiltration media was added in a 1:1 ratio, with the collection of permeate continuing until the initial protein concentration was restored. In stage 2, retentates and permeates were recycled to simulate fouling accumulation in a steady-state without altering the retentate composition. Utilizing water as the diafiltration medium resulted in higher flux and lower resistance values compared with using ultrafiltration per-meate, irrespective of the filtration mode. The concentration had a significant impact on membrane resistance, with no noticeable time-dependent effect on fouling layer development after 60 min of filtration when the retentate composition remained constant. The protein composition of the permeate and extracted foulants were comparable between the 2 media, with caseins predominating in the fouling layer.


INTRODUCTION
Micellar casein concentrates have become increasingly popular as a valuable ingredient, produced alongside native whey protein concentrates.Whey protein concentrates are known for their high nutritional value and are preferred in premium nutritional products such as infant formulas.The separation of milk into casein micelles and whey proteins is crucial to meet the desired nutritional standards for infant formula, especially in terms of amino acid composition, and to maintain consistent quality across batches.
Microfiltration (MF) has emerged as an effective technique to separate casein and whey proteins based on size, using membranes with pore sizes between 0.05 and 0.2µm (Punidadas and Rizvi, 1999).Spiral-wound polymeric membranes have become popular in the industry due to their larger membrane area per element and low cost per square meter compared with ceramic membranes.However, it has been shown that ceramic membranes are more effective in removing whey proteins from milk than spiral-wound membranes due to the higher crossflow and uniform transmembrane pressure across the membrane (Zulewska et al., 2009).However, membrane fouling remains a significant challenge in both membrane types, impacting the economic viability of the filtration process.According to existing literature, the transmission of whey protein through a membrane is subject to the formation of a protein cake layer that functions as a secondary filter, altering the transmission of soluble proteins (Jimenez-Lopez et al., 2008).The presence of such a cake layer can be monitored indirectly by measuring changes in the flux behavior and by carrying out resistance measurements.However, more work is needed to fully understand the details of this cake layer formation and the conditions affecting fouling buildup during filtration.Whey proteins and peptides are the main foulants detected when fouling is extracted from membranes used for ultrafiltration (UF) of skim milk at 50°C (Ng et al., 2018b).Different types of fouling can develop during the filtration of a suspension of dissolved particles, such as in milk.Particles can interact with the membrane surface, particle-particle interactions can occur, and a concentration polarization layer forms, forming a cake layer if a critical concentration is reached (Chew et al., 2020).This layer forms on top of the membrane surface because of filtration dynamics whereby particles and solutes get dragged toward the membrane surface and re-diffuse into the bulk phase driven by concentration gradients (Bouchoux et al., 2013).
Diafiltration during membrane filtration of milk can be applied to increase the transmission of whey protein and yield.Different diafiltration media can be added, such as water, dairy side streams, and UF permeate (Reitmaier and Kulozik, 2022).However, the effect of diafiltration on the transmission of whey protein and fouling composition has yet to be studied in detail.It has been reported that the permeate flux decreases with the use of UF permeate or a medium containing calcium compared with a diafiltration media with low or no minerals, with either ceramic membranes (Reitmaier et al., 2021) or polymeric membranes (Granger-Delacroix et al., 2023).The increased flux when water is used as diafiltration medium has been attributed to changes in the structure of the cake layer caused by increased electrostatic repulsion between the proteins caused by less shielding minerals in the serum phase (Ng et al., 2018a).
To study the impact of the reversible fouling layer in form of concentration polarization/cake layer, as well as the irreversible fouling that remain on the polymeric membrane after flushing with water, an experimental setup was created in the laboratory.The experimental setup was tailored to imitate flow conditions present in spiral-wound membranes utilized in filtration processes on a larger scale.
The objective of this research is to evaluate the influence of membrane fouling on protein transmission during microfiltration of skim milk using polymeric membranes.The study concentrates on the impact of different diafiltration media and filtration stages, specifically batch diafiltration mode (stage 1) and steady-state filtration mode (stage 2), on whey protein transmission, fouling composition, and overall filtration performance.Additionally, the research examines the effects of ultrafiltration permeate and mineral content on permeate flux, especially in varying concentration conditions during the 2 filtration stages.A comprehensive analysis of these factors will be conducted.
This study brings crucial information on how filtration parameters, such as concentration combined with diafiltration, affects this fouling mechanisms during microfiltration of skim milk and presenting a methodology to investigate the fouling mechanism in laboratory scale.

Membrane filtration
To conduct the experiment, low-heat pasteurized skim milk (72°C/15 s) was obtained from Arla Foods Amba (Denmark) and to prevent unwanted microbiological growth, 0.02% of sodium azide was added (Merck, Germany).Skim milk has been fractionated into milk protein concentrate (MPC) and UF permeate (UFP) using a Labstak M20 system (Alfa Laval, Denmark) with PES UF flatsheet membranes (MT, MWCO of 5 kDa, Synder Filtration, USA) at a temperature of 10°C and a transmembrane pressure (TMP) of 2 bar.After the fractionation process, MPC and UFP were stored overnight at 5°C.MPC was used as the initial sample as maintain the initial protein concentration of skim milk (3.5%), and to check the effect of different diafiltration media, UFP or ultrapure water were added during the experiments.
A custom-built filtration cell (EMI Twente B.V., Netherlands) was used for MF.To simulate the flow pattern close to the industrial membrane, the cell was designed to have a retentate channel height similar to the distance between 2 flatsheet in a spiral-wound membrane and a 46-mil diamond spacer was also used on top of the membrane.The filtration cell was made of 2 stainless steel plates with a PVDF MF membrane (FR, Synder Filtration, USA) with a MWCO of 800 kDa (≈0.350 µm) and the total active membrane area of 0.015 m 2 .The flow was controlled with a gear pump (LP-WT3000-1JB, Drifton, Denmark), and the pressure was adjusted using a control valve on the retentate side (Type 6013, Burkert, Denmark).The pressure before and after the filtration cell was measured using 2 inline pressure gauges (Ceramic in-line, RS PRO, Denmark), and the TMP was calculated using Equation 1.
Where P Feed = pressure measured of the feed inlet (bar), P Retentate = pressure measured of the retentate outlet (bar) and P Permeate = pressure measured of the permeate outlet (bar).In this experimental design, the P Permeate was neglected and set to 0 bar.The process was controlled, and data was collected using the Inniti software platform (Inniti, Denmark).The filtration setup is illustrated in Figure 1.A new clean MF membrane was installed in the filtration cell for each experiment.The membrane was flushed with hot tap water (40°C) for 10 min, then with demineralized water at room temperature (25°C) for 5 min before use.This was done to remove the protective layer of glycerine from the dry membrane, following the supplier's recommenda- After the membrane was prepared, the water flux was measured using demineralized water at room temperature to check for leaks or insufficient protective layer removal.The water flux was measured at a TMP of 0.5 bar and 25°C, ranging from 300 to 970 L m −2 h −1 .The extensive range in water flux can be attributed to the small membrane area used in each experiment.The pore size distribution of a membrane is not entirely uniform either based on internal analysis (data not shown), which can also give some variation in the measured water flux.
The feed was run through the cell at 50°C for 5 min with a crossflow velocity (CFV) of 0.23 m/s.The restriction valve was fully open, and the permeate and retentate were recirculated back to the feed.Hartinger et al., (2020) found that increasing CFV in the range of 0.15-0.33m/s during microfiltration of skim milk using polymeric membranes increased the flux and the whey protein transmission (Hartinger et al., 2020).A CFV of 0.23 m/s was chosen to simulate the flows used for MF of skim milk with polymeric spiral-wound membranes, which could be handled in a laboratory setup without too much foam formation of the milk.
The crossflow velocity (CFV) is the flow tangential to the membrane surface and is calculated using Equation 2.

CFV Flow rate Crosssectional area
where the flow rate is the volumetric flow of the feed (m 3 s −1 ) in the flow channel, and the cross-sectional area (m 2 ) is the area of the cross-section of the flow channel.The flow rate used in this study was 1.64 × 10 −5 m 3 s −1 , and the cross-sectional area (width x height of channel) was 7.01 × 10 −5 m 2 .During the MF experiment, the diafiltration step was carried out by adding a diafiltration medium to the MPC (produced by UF) as UFP or ultrapure water (Milli-Q) at a 1:1 volume ratio.
UFP which has been collected during UF of skim milk, was chosen as it maintains the natural environment for the casein micelles and consists mainly of lactose and minerals.Ultrapure water, conversely contains low levels of minerals and will modify the serum phase (Coşkun et al., 2022).Nevertheless, water is often used as a diafiltration medium in the industry, such as water produced by reverse osmosis process, which typically has as low content of minerals as ultrapure water.The feed was heated to 50°C and subjected to the pre-conditioned membrane as discussed previously.

Filtration modes
The experiment involved 2 stages of filtration: batch diafiltration mode (stage 1) and steady-state filtration mode (stage 2).In the batch diafiltration mode (stage 1), the original 2x MPC was diluted with UFP or water at a 1:1 ratio, followed by concentration to achieve the original protein content.The filtration parameters were kept constant at a temperature of 50°C, TMP of 0.5 bar, and a crossflow velocity of 0.23 m/s.In stage 1, the retentate was recirculated, and the permeate was removed as a batch concentration step.The permeate flux and pressure were constantly monitored.At the end of the process, the membrane was flushed with demineralized water to measure water flux at 25°C.The membrane was then stored at 5°C for further analysis.The permeate and retentate samples were collected for analysis when the concentration factor of 2x was reached.The retentate was used as feed in the second MF experiment process (stage 2).Trial Settings for stage 2 were the same as for stage 1, but both permeate and retentate were recirculated to keep the concentration level to a minimum.Samples of the permeate were collected at 5, 10, 30, and 60 min to follow protein composition of the permeates over time.A feed volume of 200 g was used for each experiment, and due to ongoing permeate sampling during the stage 2 experiment, the total experiment resulted in a feed concentration of approximately 25%.The experiment was repeated the same, but this time without sampling during filtration to evaluate the potential impact of feed concentration resulting from the permeate sampling.After the last permeate sample was collected, the membrane was flushed, and water flux was measured at 25°C.The retentate, permeate, and membrane samples were analyzed.A flowchart of all the filtration steps is shown in Figure 2.

Resistance measurements
Equation 3 can be used to calculate the resistance of a membrane after measuring the water flux of the new membrane.The following formula can be used: where R m = membrane resistance (m −1 ), TMP is the difference in the transmembrane pressure (kg m −1 s −2 ), J is flux (kg m −2 s −1 ), η is the kinematic viscosity (m 2 s −1 ) of water.
To calculate the total resistance experienced during the filtration experiment, it is necessary to take into account both the resistance of the irreversible fouling, which requires cleaning to be removed, and the reversible foul-ing, which consists of the concentration polarization and cake layer.Equation 4 demonstrates how to calculate this total resistance using the permeate flux In this equation, R irr stands for resistance from the irreversible fouling, that needs cleaning before it can be removed (m −1 ), R rev refers to the resistance from the reversible fouling, and η is the kinematic viscosity (m 2 s-1 ) of the MF permeate.
To calculate the resistance from the irreversible fouling (R irr ), measure the water flux after filtration as shown in Equation 5: R R TMP J m i rr water The dynamic viscosity of the MF permeates is measured using a Rheometer with a double gap cell (Physica MCR 301, Anton Paar, Austria), with water added as diafiltra- An overview of the filtration processes is as follows: The initial ultrafiltration (UF) process is carried out at 10°C, resulting in the production of a permeate (UFP) and a retentate (UFR).The microfiltration (MF) is conducted in 2 stages at 50°C, with a new membrane installed at each stage.In the first stage (stage 1), water or permeate is added at the beginning, and the retentate is recirculated.In the second stage (stage 2), both the permeate and retentate are continuously recirculated to observe potential differences in the buildup of the cake layer with or without concentration.Dashed lines are used to illustrate when the streams are being recirculated.The retentate stream is referred to as UFR/MFR and the permeate stream is referred to as UFP/MFP.tion medium and UFP added as diafiltration medium at 50°C and 204 s −1 .The dynamic viscosity of the MF permeates is 0.623 × 10 −3 Pa s with water added as a diafiltration medium and 0.655 × 10 −3 Pa s with UFP added as a diafiltration medium.The densities of the MF permeate are 1.004 g cm −3 (water) and 1.013 g cm −3 (UFP) at 50°C.Water's dynamic viscosity and density at 20°C (1.002 Pa s and 0.998 g cm −3 , respectively) are used to calculate R m and R f (Kestin et al., 1978).

Extraction of foulants from MF membranes
To investigate the fouling left on the membrane after flushing the cell with water, the MF membrane was removed from the filtration cell after each experiment and cut into 2 pieces of 12x5 cm.The membrane piece facing the front end of the cell was soaked in 50g of 2M Urea (≥99.0%Urea GR, Merck, UK) and 0.1M Trisodium Citrate solution (≥99.0%EMSURE,Merck, Germany).The extraction solvent was chosen to maximize the extraction and solubility of the proteins to be analyzed by reversephase UPLC (RP-UPLC) (Ng et al., 2018b).Preliminary studies showed that 2M Urea resulted in similar band intensities as extracts with 1% SDS and that both extract methods mainly consisted of caseins (data not shown).The membrane and solution were shaken for 24 h at room temperature, and the extracts were analyzed for total protein, minerals, SDS-PAGE, and RP-UPLC.

Sodium dodecyl sulfate-polyacrylamide-gel electrophoresis (SDS-PAGE)
Permeate samples and extracts were analyzed using SDS-PAGE.Permeate samples were diluted to 0.25% protein in UF permeate.Extracts were undiluted due to the very low protein concentration, as analyzed by Kjeldahl (<0.05% total protein).
The permeate samples were analyzed in reduced and non-reduced conditions.For non-reduced samples, 5µL of the sample was mixed with 5µL 2x Laemmli sample buffer (Bio-Rad, Denmark).For the reduced samples, 5µL of the sample was mixed with 4.5µL 2x Laemmli sample buffer and 0.5 µL 1M Dithiothreitol (DTT) (≥98%, Sigma-Aldrich, Denmark).Samples were heated at 90°C for 10 min in a heating block.
After the heating step, a precast gel was loaded with 10 µL of sample in each well (Mini-Protean®TGXGel, Any kD, Bio-Rad, Denmark).5µL of the standard marker was added to each gel (10-250 kD, Bio-Rad, Denmark).One x Tris/Glycine/SDS was used as the running buffer (10xTGS, Bio-Rad, Denmark), and electrophoresis was performed with a constant voltage of 150V.The gel was then washed with demineralized water and stained with 50 mL Coomassie blue for 1 h (Bio-Rad, Denmark).The gels were de-stained overnight with demineralized water.

Reversed-Phase ultra-high-performance chromatography (RP-UPLC) of milk proteins
RP-UPLC was applied to analyze the relative changes in the composition of milk proteins of the feed, the MF permeates, and extract samples using an adjusted method described by Bonfatti et al. (2008).This was used to follow if the overall composition of the permeates and hence evaluate the protein transmission.Skim milk was used as a reference to identify peak elution times.Permeate samples were diluted 2 times with a reduction buffer consisting of 6M urea (≥99.0%Urea GR, Merck, UK), 0.1 M Trisodium Citrate dihydrate (≥99,0%EMSURE, Merck, Germany), and 0.02M DTT (≥98%, Sigma-Aldrich, Denmark).The extract samples were not diluted as they already contained urea and Trisodium Citrate buffer.The diluted samples were mixed for 60 min.After passing the samples through a 0.22 µm cellulose acetate filter (VWR, Denmark).Ten µL of the sample was injected into the RP-UPLC (Acquity H-Class UPLC with a UV 214nm detector, Waters, Denmark).A C18 column was used as the analytical column in this experiment (Poroshell 120 SB, Agilent, Denmark).Eluent A consisted of 0.1% Trifluoroacetic acid (TFA) (≥99.8,Merck, Denmark) in Milli-Q water, and eluent B consisted of 0.1%TFA in acetonitrile solution (≥99.5% Rathburn Chemicals, Mikrolab, Denmark).The separation was conducted at 42°C with a flow rate of 0.35 mL/min.The following gradient was applied: 31.8-43.5% of eluent B over 15.5 min, followed by 43.5-100% of eluent B over 5 min.Subsequently, a gradient of 0-100% of eluent B was applied every 2 min until a total run time of 41.5 min was reached.Finally, the eluent B composition was set at 100-31.8% for 5 min.

Zeta potential of membranes
To investigate any differences between a new and fouled membrane, the Surpass 3 surface analyzer with an adjustable gap cell (Surpass 3, Anton Paar, Austria) was used to measure their streaming potential.The zeta To prepare the membranes for testing, they were cut into small pieces (20x10 mm) from the middle and attached to the sample holder.A 0.001M KCl solution was then used as the electrolyte solution for resistance measurements.After adjusting the gap height to 110-120 µm, a pH scan was conducted from around 5.5 to 3 in intervals of 0.3 using 0.1 M HCl (Sigma Aldrich, Denmark).Each pH scan included 3 streaming potential measurements and 3 rinse cycles with the electrolyte solution.

Statistical analysis
The data analysis involved a 2-way ANOVA with diafiltration media and filtration time as independent variables.A one-way ANOVA was performed to compare the final composition of the feed, retentate, and permeate samples subjected to different diafiltration media.Tukey-Kramer's test was then applied as a post hoc test to identify the significant differences between the samples.A significance level of P = 0.05 was used for this study.
The data analysis package of Excel® Microsoft® 365 was utilized for the ANOVA analysis.Filtration trials with UFP as a diafiltration medium were repeated 4 times, while water was used as a diafiltration medium 7 times over 4 independent weeks.To evaluate the setup and analyses' repeatability, a replication of the filtration trial with water as a diafiltration medium was conducted in 3 of the weeks.
Regression lines and bar plots were created, and standard deviationa and mean values were calculated using Jupyter Notebook and Python Software Foundation, Python 3, https: / / www .python.org/ .

Filtration performance of milk protein concentrate (MPC) in 2 different filtration modes
The chemical composition of skim milk, milk protein concentrate (MPC), and ultrafiltration permeate (UFP) is shown in Table 1.The MPC contains higher levels of calcium, phosphorus, and magnesium compared with the initial skim milk and the UFP.This is expected as the casein micelle contains 65% of the calcium, 40% of the inorganic phosphorus, and 30% of the magnesium which exist in skim milk.The remaining amount is in a diffusible form (Holt, 2004).The UFP does not contain any proteins but has a small concentration of non-protein nitrogen (NPN) of 0.03%.Compared with skim milk, the UFP has lower concentrations of calcium, phosphorus, and magnesium as it represents the soluble phase of skim milk.
The filtration performance of the MPC was measured by following the permeate flux over time with different diafiltration media for the 2 stages of filtration (Figure 3).As mentioned in section 2.1.1,after the first stage, the retentates from stage 1 were collected and used as feed for the stage 2 filtration after installing a new membrane in the filtration cell (Figure 2).By using the retentate from stage 1 concentration, the fouling buildup could be followed at a concentration factor of 2x, with different mineral backgrounds from the previous diafiltration step.By recirculating the retentate and permeate, changes in retentate viscosity affecting the permeate flux would be minimized.
The flux values were significantly lower when diafiltration was carried out with UFP compared with water for stage 1 (Figure 3A) and stage 2 (Figure 3B) filtration.According to a recent study utilizing the same type of membrane, flux was compared during the diafiltration of skim milk with UFP and water at 10°C (Granger-Delacroix et al., 2023).The authors found that the flux increased over time when water was used as a diafiltration medium, while the flux was lower and constant using UFP as diafiltration medium, also when adjusted for viscosity.
Flux showed a significant decline over time when the feed was concentrated during stage 1 (Figure 3A).Due to the lower permeate flux when UFP was added in stage   1 filtration mode, a longer filtration time of around 10 min was needed to remove the same amount of permeate and obtain the same concentration factor as when water was used.In the present experiment, the filtration lasted a maximum of 60 min, and the amount of milk flow through the membrane surface ranged between 45 and 60 L, which was considered sufficient for building up a representative fouling layer.
In the stage 2 filtration mode, the flux also declined over time, which was hypothesized to be due to sampling during the filtration (Figure 3B).When the stage 2 filtration was repeated without sampling and consequently no concentration, the flux, during the experiment, remained the same as the initial flux (Figure 3B, filled triangles).Therefore, it was concluded that the significant decrease of flux in stage 2 was due to the increase in feed concentration during sampling and not because of alteration or buildup of the fouling layer over the 60-min processing time.
Figure 4 displays the resistance of the reversible fouling (R rev ) over filtration time, in addition to the permeate flux, for both stage 1 and stage 2 filtration.Notably, UFP exhibited a significantly higher resistance compared with water in stage 2 during the steady stage, both with and without sampling, which aligns with observation in other literature (Reitmaier et al., 2021).The resistance values were determined by taking the differences in permeate viscosity at different diafiltration media into account.This suggests that the reversible fouling were affected by the diafiltration medium, but only at particular concentration levels where fouling buildup occurred.As a result, resistance increased.It appears that it was at these crucial concentrations where the diafiltration media had an impact on the layer.
No difference was found in the irreversible fouling between the 2 stages or the diafiltration media, as shown in Figure 5.It can also be noted that the irreversible fouling contributes only a small portion to the total resistance.

Composition of permeates produced from stage 1 or stage 2 filtration with different diafiltration media
Mineral composition of permeates.The mineral composition of the permeates from stage 1 and stage 2 filtration treatments is presented in Table 2, including calcium, phosphorus, magnesium, potassium, and sodium.As anticipated, the permeate produced using water as the diafiltration medium exhibited a considerably reduced mineral content, owing to the minerals in the original serum phase being replaced by water.Research indicates that only a fraction of the minerals found in milk, including calcium, phosphorus, and magnesium, exist in a diffusible form (30-50%), with the remainder being linked to the casein micelle (Holt, 2004).
No significant differences in mineral levels were observed between filtration modes, except for lower phosphorus levels during stage 2.This difference was consistent regardless of the diafiltration media used and could be attributed to a lower permeation of soluble casein aggregates rich in phosphorus during stage 2 filtration.The reduced permeation of soluble casein aggregates in stage 2 could be explained by their prior removal from the permeate during the concentration at stage 1.Moreover, the most significant decrease in phosphorus was observed from stage 1 to stage 2 when water was used as the diafiltration medium.This could be attributed to an increased resistance of the fouling layer during stage 2 filtration mode as seen in Figure 4B, hindering the transmission of soluble casein aggregates through the membrane.
Other studies have reported increased levels of soluble calcium and phosphate when skim milk was concentrated to 4x by MF at 10°C after diafiltration with water.This was explained by calcium and phosphate permeating from the casein micelle (Coşkun et al., 2022).The mineral profile for the retentates can be found in the Supplementary data (Table S1).
Protein transmission during filtration.In this study, differences in the reversible fouling layer and the effect on protein transmission were investigated using UFP and water as diafiltration media.The protein composition of the permeates was measured after stage 1 and 2. It was postulated that the observed rise in filtration resistance when utilizing UFP as the diafiltration medium was caused by the development of a reversible fouling layer with greater resistance, in contrast to the use of water, which could have an impact on protein transmission.Table 2.The mineral content of collected permeate from the first filtration (stage 1) or permeate sample after 60 min of filtration from the second filtration (stage 2), using permeate (UFP) or water as diafiltration media.Mean ± standard deviation shown for n ≥ 3. Values having different subscript letters within a column are significantly different (P < 0.05) pH [-] Na [mg kg In a study by Hartinger et al. (2019), reversible fouling in skim milk during MF was examined.According to the authors, increasing the TMP leads to compression of the layer, resulting in reduced transmission of casein and whey proteins.This is thought to be primarily due to the formation of a denser and more compact cake layer, as evidenced by lower flux and higher resistance data (Hartinger et al., 2019).
The permeates in this study contained mainly whey proteins and traces of caseins.The permeates did not contain α s1 -casein, but κ-casein, α s2 -casein, and β-casein were present (Table 3).There were no differences in the relative casein composition between diafiltration media or filtration mode, neither for casein nor for whey proteins.Hence, although the fouling layer when UFP was used as a diafiltration medium gave a higher resistance, it did not affect the overall protein composition.
When no sampling was carried out during stage 2, maintaining the concentration constant throughout the filtration time, the same protein composition was measured for the permeate samples, except for a significantly lower β-casein/total casein when water was used as a diafiltration medium (36.7% with UFP and 31.2% ± 1.13 with water).
A recent study investigated the release of calcium and casein into the serum phase by dialysis of suspensions of micellar casein powder in deionized water or simulated UF permeate (SMUF) at 2 different temperatures, 10°C and 50°C (Reitmaier et al., 2020).During dialysis in deionized water at 50°C, an increased release of α s1 -casein, κ-casein, and β-casein into the serum phase was found, as well as a decrease in ionic calcium compared with the re-suspensions in SMUF.No change was found when the micellar casein powder was suspended in SMUF.The release of caseins from the casein micelle into the serum phase can be explained by the lower ionic calcium and higher pH when deionized water was used.
According to a recent study by Coşkun et al. (2022), it was noted that during the 2x concentration of skim milk at 10°C, there was an increase in α s1 -casein and a decrease in α s2 -casein/total casein area in the soluble phase of skim milk and retentates.This was achieved by using a vibrating filtration cell to minimize the impact of fouling layer formation.There was also a decrease of β-casein/ total casein and an increase of κ-casein/total casein of the retentates during the concentration step.However, these values remained constant during the following diafiltration step with water.The release of κ-casein and α s1casein into the serum phase was explained by changes in the stabilizing layer and interparticle interactions occurring during concentration (Coşkun et al., 2022).
In this study, no α s1 -casein was found in the permeate samples, but a higher κ-casein and α s2 -cas/total casein was found compared with MPC (Table 3).The discrepancy between the present data and those reported before could be possibly due to the differences in the operating temperature and/or type of filtration.For instance, a recent study showed that the mode of diafiltration (continuous vs batch) affects the formation of soluble casein aggregates in milk during cold MF, particularly when water is used as diafiltration media (Raak et al., 2023).However, a more thorough study is needed to assess this further.
Although minor differences were found in the overall protein composition of the permeates, the release of caseins into the serum phase could greatly impact the formation of the fouling/cake layer by affecting the structure or forming soluble complexes that can be retained.

Protein composition of extracted foulants
The composition of the fouling layer was analyzed using SDS-PAGE and RP-UPLC.SDS-PAGE showed that there were no apparent differences in the polypeptides' migration.Regardless of the filtration mode or diafiltration media used, all extracted foulants had similar compositions with caseins being the main foulant (Supplementary data).A more detailed RP-UPLC analysis revealed that β-casein was the highest intensity peak regardless of the diafiltration media or filtration mode.In contrast, peaks  3. Protein content (g kg −1 ) and relative protein composition (% of total area of caseins (CN) or whey proteins) of MPC and permeates after the first (stage 1) and second (stage 2) filtration stages using permeate (UFP) or water as diafiltration media.Mean ± standard deviation shown for n ≥ 3. Values having different subscript letters within a column are significant different (P < 0.05) with lower intensities were found for κ-, α s1 -, and α s2casein (Figure 6).Interestingly, only a small amount of whey proteins was found on the membrane, and less than 10% of the total area in the chromatogram was due to the presence of whey proteins (Figure 6).These results contradict previously published reports, where whey proteins, and not caseins, were found in the extracted foulants of PES UF membranes after filtration at 50°C (Ng et al., 2018b).The discrepancy may be due to the smaller pore size of a UF membrane and/or the difference in membrane material compared with the membrane used in the present study (PES vs PVDF).Indeed, PVDF is more hydrophobic than PES and has been found to adsorb more peptides, when soaked in a whey protein hydrolysate solution compared with PES (Kadel et al., 2019).
Characterizing the fouling left on the membrane after flushing with water is of great interest from the industry, while this can be used to optimize the cleaning procedures.For instance, primarily protein was found as the foulants, and hence the cleaning procedure should be targeting removal of proteins as the first step.This could potential reduce the amount of cleaning steps, and water used for flushing in between the steps, which would be more economical for the dairy plant and sustainable in regards of water saving.

Surface modifications of fouled membranes
After filtration, the membranes were rinsed with water, and their zeta potential was measured.The results were then compared with the zeta potential of the initial membrane.The new membranes showed a negative zeta potential in the pH range of 5.7-3, as shown in Figure 7.However, the fouled membrane showed a net charge of 0 mV at around pH 4.4-4.6 and reached a net positive charge of +26 mV at pH 2.9.No significant differences were found between the different diafiltration media or filtration modes.These zeta potential values measured on the fouled membrane supported the findings that the proteins present in the extracts were mainly caseins.This is because the zeta potential of micellar casein is around +30 mV at pH 3, which means they have a positive net charge at pH 4.6, the isoelectric point of the casein micelle (Post et al., 2012).The same study also showed that the positive net charge of β-casein can be found at pH 4.8.

CONCLUSION
In conclusion, this study found that substituting the serum phase of milk with either water or UFP during diafiltration can modify permeate flux and the resistance of the reversible fouling layer.The concentration of the feed negatively impacts the flux and resistance of the reversible fouling layer more than the time effect on the fouling layer buildup using polymeric membranes.However, choosing the diafiltration media and filtration mode does not affect the relative protein composition of the MF permeate.The differences in filtration performance between the 2 diafiltration media may be attributed to altered mineral profiles in the serum phase.Additional research is required to further explore the underlying factors behind these findings.This requires investigating whether the modification in filtration efficacy is a consequence of the development of soluble aggregates that affect filtration resistance by manipulating the reversible fouling layer.Moreover, it could be attributed to differences in electrostatic repulsions between the proteins within the reversible fouling layer when the mineral concentration of milk's serum phase is partly replaced with water.
Mardal et al.: Impact of Diafiltration… tions.No cleaning agent was used to avoid altering the membrane surface.

Figure 1 .
Figure 1.Illustration of filtration system used for the filtration experiments.Permeate can be removed or added back to the feed tank depending on the filtration mode.

Figure 3 .
Figure 3. Permeate flux (L m −2 h −1 ) as a function of filtration time during the first (stage 1, A) and second (stage 2, B) stages, while diafiltering with water (•) or UFP (X).Filled triangles represent the values from stage 2 with no sampling during the filtration (▼UFP/ ▼Water).Error bars show the standard deviation for each sample.

Figure 4 .
Figure 4. Resistance of reversible fouling (R rev ) shown during microfiltration with water (•) or UFP (X) added as diafiltration media as a function of filtration time.A) Stage 1 or B) stage 2 filtration.Filled triangles represent the values from stage 2 with no sampling during the filtration (▼UFP/ ▼Water).Error bars show the standard deviation for each sample.

Figure 5 .
Figure 5.The data on irreversible fouling resistance (Rirr) is shown from membrane samples collected after each filtration trial and flushing with water.The bars represent the diafiltration media, either UFP (black) or water (gray), used in the first or second stage of filtration.The error bars indicate the standard deviation for each sample.

Figure 7 .
Figure 7. Zeta potential as a function of pH for a new membrane (black solid line -X) and after their use after diafiltration with permeate (black solid line -•) or water (gray dashed line -▼).The error bars show the standard deviation of the measurement.

Table 1 .
Composition (protein, dry matter and selected ions) for skim milk, milk protein concentrate (MPC) and UF permeate.Mean value with standard deviation shown for each component.Values having different subscript letters within a column are significant different (P < 0.05)