1. Introduction
The biotechnological processing of agro-industrial residues enables the valorization of waste streams. There is global interest in the use of such residues as a raw material for the production of fuels, biological preservatives, pharmaceuticals, and other high-value goods, but there is still a need for more research [
1]. Bacteria and yeasts can be used as biocatalysts for a wide range of applications involving the conversion of residues into high-value products. The yeast
Kluyveromyces lactis is particularly suitable for this task because it can utilize diverse nutrient sources, produce large quantities of recombinant proteins, and has a generally regarded as safe (GRAS) status [
2].
Following fermentation, the liquid and solid components of the broth must be separated. Centrifugation, depth filtration or membrane filtration are suitable for this separation task, but membrane filtration is the least expensive in terms of upfront costs and the most scalable. Additionally, sterile membrane filtration is mandatory when using genetically modified organisms to ensure the removal of cells [
3]. However, one disadvantage of membranes is the high cost of consumables, so it is important to reuse the membranes as much as possible [
4]. With a life span of more than 10 years and the ability to tolerate harsh chemicals and extreme pH, ceramic membranes are ideal for a long service life [
5,
6]. The reusability of such membranes requires us to address the challenges of flux decline due to fouling and unavoidable cleaning, thus restoring membrane permeability.
A comprehensive review of cleaning methods for ceramic membranes was recently published [
7]. The susceptibility of membranes to fouling (and thus the loss of flux) is influenced by the feed composition, hydrodynamic conditions, and membrane properties. The latter include surface roughness, the thickness of the separation layer, charge, hydrophobicity, and functional groups on the surface [
8]. The surface roughness is particularly relevant when the size of the filtered particles or colloids is similar to or smaller than the asperities on the surface [
9,
10]. When particles or colloids are smaller than the asperities, the probability of deposition in the valleys increases [
11]. However, the filtration resistance of a membrane increases as the separating layers become thicker [
12,
13].
Fouling analysis is required for the characterization of filtration processes. This includes detecting the onset of fouling and the analysis of factors related to fouling, helping to explain fouling mechanisms and the quantification or prediction of fouling behavior [
14]. For example, the biofouling potential of industrial fermentation broth during microfiltration (MF) was investigated by scanning electron microscopy (SEM) and scanning acoustic microscopy (SAM), which allow the qualitative determination of fouling mechanisms, together with the resistance-in-series (RIS) and combined pore-blockage and cake filtration models, which facilitate quantitative analysis [
15]. The fouling of MF [
16], ultrafiltration (UF) [
17], and reverse osmosis (RO) [
18] membranes, among others, has been analyzed qualitatively by SEM, whereas RIS models are used regularly for the quantitative analysis of fouling involving biological foulants such as enzymes [
19], protein aggregates [
20], viruses [
21], and bacteria [
22].
The factors that affect membrane fouling, including surface roughness and separation layer thickness, are often investigated using model solutions such as bovine serum albumin (BSA) [
23]. For the separation of yeast biomass, most studies use baker’s yeast (
Saccharomyces cerevisiae), typically freeze-dried and rehydrated even though the results tend to differ compared to freshly cultivated cells [
16,
24]. Here we considered the industrial strain
Kluyveromyces lactis and studied the influence of membrane surface properties on fouling during the separation of biomass from the fermentation broth. The cells were cultivated in a medium derived from agro-industrial residues containing corn steep powder and whey and were filtered using ceramic membranes with different pore sizes (30 nm and 200 nm) and geometries (tubular and hollow fiber). Here, we focused on analyzing the surface roughness and the composition of the separation layer for two 30 nm cut-off tubular membranes and determined their individual susceptibility to fouling by proteins and yeast cells. We also investigated the influence of operational parameters such as crossflow velocity (CFV) on filtration efficiency and fouling.
2. Materials and Methods
2.1. Strain and Fermentation Medium
For biomass production, we used K. lactis strain GG799 (New England Biolabs, Frankfurt, Germany). The fermentation medium consisted of 3.0% (w/v) corn steep powder SOLULYS 095E (Roquette Frères, Lestrem, France), 0.6% (w/v) heated and crystallized sweet whey powder (Bayrische Milchindustrie, Landshut, Germany) and 150 mM citrate-phosphate buffer (pH 5), comprising 10.2 g L−1 citric acid monohydrate (AppliChem, Darmstadt, Germany) and 14.6 g L−1 disodium hydrogen phosphate (Merck, Darmstadt, Germany). To reduce the natural particle load, the suspensions of corn steep and whey powder were passed through a 100 nm Al2O3 filter before use.
2.2. Cultivation of K. lactis
We used a 5-L bioreactor system (Applikon Biotechnology, Delft, The Netherlands) with a working volume of 4 L for the cultivation of K. lactis. The bioreactor was operated at 30 °C, with an air aeration rate of 3 L min−1 and an agitation speed of 800 rpm. The pH was not controlled. To avoid foaming, we added 0.01% (v/v) Struktol J673A (Schill & Seilacher, Hamburg, Germany) before inoculation. The culture was inoculated with a glycerol cryo-stock to an optical density of ΔOD600 = 0.1.
2.3. Harvest of Medium and Biomass
The culture was harvested after 25 h, having reached the stationary phase. The fermentation broth was drawn aseptically and cooled immediately in an ice bath. The biomass and the medium were separated by centrifugation (17,207× g for 10 min at 4 °C) in a Sigma 6–16 KS centrifuge equipped with an 11,650 rotor and six 13,650 cups (Sigma Laborzentrifugen, Osterode am Harz, Germany). The medium was passed through a 0.22-µm polyethersulfone filter (Corning, New York, NY, USA) and stored at 4 °C. To remove residual medium, the biomass pellet was resuspended in 154 mM NaCl and centrifuged as described above. The supernatant was discarded, and the pellet was resuspended in 150 mM citrate-phosphate buffer (pH 5) and stored at 4 °C.
2.4. Filtration Setup
The technical data for the ceramic membranes used in this study are summarized in
Table 1. All membranes were manufactured and kindly provided by MANN+HUMMEL (Ludwigsburg, Germany), except for the monochannel 200 nm MF membrane, which was purchased from Atech Innovations (Gladbeck, Germany). The feed solution (1.5 L) was circulated using an FCPA 80B-4/HE rotary vane pump (AFT, Rosstal, Germany). The temperature of the feed vessel was measured using a PT-100 sensor and was maintained at 25 ± 1 °C by immersion in a water bath. The transmembrane pressure (TMP) was adjusted using a manual ball valve located behind the membrane module and was measured using two type-401001 sensors (JUMO Mess- und Regeltechnik, Vienna, Austria) in front and behind the membrane module. The feed volume flow was measured using an SM6000 magnetic-inductive flow sensor (ifm electronic, Essen, Germany). The permeate flow was measured gravimetrically using a DS 8K.05 balance (Kern & Sohn, Balingen, Germany). Filtrations were carried out in total recycle mode (TRM), and the permeate was regularly recycled to the feed. Data were recorded using a LABmanager 1 and the corresponding software LabVision v2.9 (both from HiTec Zang, Herzogenrath, Germany). The filtration setup is shown in
Figure 1.
2.5. Estimation of Intrinsic Membrane Resistance and Irreversible Resistance
The intrinsic membrane resistance R
m [m
−1] was determined before filtration and after chemical cleaning. It was calculated from the pure water flux at a CFV of 0.8 m s
−1 (1.6 m s
−1 for the 9C HF membrane) and TMPs of 0.1–1.1 bar at 25 °C using Equation (1):
where R
m is the intrinsic membrane resistance [m
−1],
η is the permeate dynamic viscosity [Pa s], TMP is the transmembrane pressure during filtration [Pa], and J
w is the pure water permeate flux [m
3 m
−2 s
−1]. The viscosity of water is shown in
Table 2. To calculate the irreversible fouling resistance R
n, irrev (Equation (2)), the fouled membrane was flushed with deionized water at 50 °C for 20 min using the same process parameters described above and with the permeate valve closed, then with fresh deionized water at 50 °C for 20 min under the same process parameters with the permeate valve open. The index n indicates the fouling resistance of the fermentation broth, medium, or yeast cells, as appropriate.
The pure water flux of the rinsed membrane was then measured again as described above. The irreversible resistance corresponds to the resistance that can only be removed by chemical cleaning and not water rinsing.
2.6. Investigation of Total, Fouling, and Reversible Resistance
We determined the total resistance R
ferm, tot of each membrane by filtrations with fermentation broth. The feed consisted of 0.75 L medium and biomass to a concentration of 2.5 g L
−1 cell dry weight (CDW) mixed with 150 mM citrate-phosphate buffer (pH 5) in a total volume of 1.5 L. When the feed was fully mixed, the pH was adjusted to pH 5 with 37% hydrochloric acid (Carl Roth, Karlsruhe, Germany). Before filtration, the feed was circulated in the system for 10 min at 0.5 bar TMP, a CFV of 0.8 m s
−1 (1.6 m s
−1 for the 9C_HF membrane) and a temperature of 25 °C with the permeate valve closed. The valve was then opened, and filtration was carried out with manual back-recycling of permeate, until a steady-state flux was established. The total resistance R
ferm, tot and the fouling resistance R
ferm were calculated using Equation (3). Single-component resistances were investigated using the 30 nm tubular UF membranes 7C_s and 7C_r. Feed preparation and filtration were then carried out using medium or biomass as described above, and the corresponding resistances R
yeast and R
medium were calculated using Equation (3).
where J
ss is the respective steady-state flux, and the index n indicates the fouling resistance of the fermentation broth, medium, or yeast cells, as appropriate. The respective viscosities are shown in
Table 2.
The reversible resistance was calculated by subtracting the irreversible resistance from the fouling resistance using Equation (4).
2.7. Chemical Cleaning
To restore the initial filtration performance of each membrane after estimating the irreversible resistance, chemical cleaning was carried out with 1% (w/v) P3 Ultrasil 14 (Ecolab Deutschland, Monheim am Rhein, Germany) at 60 °C, 0.5 bar TMP, and a CFV of 0.8 m s−1 (1.6 m s−1 for the 9C HF membrane) for 2 h with the permeate valve open. The system was then rinsed with deionized water until the permeate returned to neutral.
2.8. Reynolds Number and Wall Shear Stress Calculations
To characterize the flow regime, we calculated the Reynolds number (Re) using Equation (5):
where
ρ is the density [kg m
−3], CFV is the crossflow velocity [m s
−1], d is the inner channel diameter [m], and
η is the dynamic feed viscosity [Pa s]. The wall shear stress
τw [Pa] was calculated using Equation (6):
where
λ is the drag coefficient calculated as a function of the Reynolds number using the Blasius correlation (Equation (7)). This is applied in the presence of turbulent flow and hydraulic smooth pipes (2320 < Re < 10
5).
When the roughness of the surface is higher, the drag coefficient is calculated for the transition area using the Colebrook equation. However, the roughness encountered in this study was below the transition area [
25].
2.9. Viscosity Measurement
Viscosity was determined using a Haake RS 300 rheometer (Thermo Fisher Scientific, Waltham, MA, USA) fitted with a plate-cone measuring device (2° angle) equipped with a Haake DC30 thermostat (Thermo Fisher Scientific) for temperature control.
2.10. Particle Size Measurements
The particle size distribution of K. lactis cells after harvest was measured in the range 0.02–2000 µm with a Mastersizer 2000 (Malvern Panalytical, Malvern, UK) using the laser diffraction method. For each sample, ten measurements were taken with automatic parameter calculation.
2.11. Measurement of Biomass Concentration
Biomass concentrations were determined by measuring the optical density ΔOD600 and CDW. Absolute OD600 values were measured with a BioSpectrometer basic (Eppendorf, Hamburg, Germany). The samples were diluted with 150 mM citrate-phosphate buffer (pH 5) to OD600 < 0.5. The ΔOD600 was calculated from the difference between sample absorbance and blank, multiplied by the dilution factor. The CDW was determined by weighing the washed pellet from 2 mL samples after drying for 24 h at 80 °C.
2.12. Bradford Assay
Total protein was measured using Bradford reagent prepared from 0.1 g L
−1 Coomassie Brilliant Blue (Applichem), 47.5 mL L
−1 ethanol (Applichem), and 136 mL L
−1 75% orthophosphoric acid (VWR, Radnor, PA, USA). BSA (Carl Roth) was used as a standard. For the colorimetric assay, 30 µL of sample, blank or standard, was mixed with 270 µL of Bradford reagent and incubated for 5 min. The absorbance was measured at 450 and 590 nm using a Synergy HT plate reader (BioTek, Winooski, VT, USA). The absorbance ratio 590/450 nm was used for quantification [
26].
2.13. Surface Characterization of Membranes 7C_s and 7C_r
Membrane surface roughness was estimated by the manufacturer (MANN+HUMMEL) according to standard operating procedures using a VK-X110 3D laser scanning confocal microscope and the associated software v2.8.0.0 (both from Keyence, Neu-Isenburg, Germany). The 50× magnification data were processed by locally estimated scatterplot smoothing (LOESS) using OriginPro 2019b (Origin Lab, Northampton, MA, USA). The resulting data were used to calculate the arithmetic mean roughness parameter R
a using Equation (8):
where Z
j is the current value at the measurement point, and N is the number of measurement points. The thickness of the separation layers was determined by the manufacturer (MANN+HUMMEL) according to standard operating procedures again using a Keyence VK-X110 3D laser scanning confocal microscope. Briefly, the membrane was broken into fragments and images of the interface were captured at 50× magnification. The thickness of each layer was calculated using Keyence software v2.8.0.0.
2.14. SEM Images for Fouling Analysis
Fouling of the separating layers after filtration was characterized using an EVO LS 10 scanning electron microscope (Carl Zeiss, Oberkochen, Germany) in high-vacuum mode. After filtration, the membranes 7C_s and 7C_r (45 mm in length) were rinsed with water and then broken into fragments and dried for 24 h at 50 °C. Before analysis, the surface was sputtered with gold to prevent surface charge build-ups.
4. Discussion
MF membranes with pore sizes of 0.1–0.2 µm are often used to separate cells from culture medium. However, particles that are close to the pore size are likely to access the pores, leading to a severe fouling.
S. cerevisiae (cell size ≤ 10 µm) is often used as a model for membrane filtration, but
K. lactis is smaller [
27], and our data confirmed a size distribution of approximately 2–5 µm. We therefore focused on the investigation of fouling on the surface of 30 nm membranes during the filtration of medium containing
K. lactis cells. The water permeability of membranes is determined by the pore size, overall porosity, and the thickness of the separation layers [
28]. In our experiments, the membrane with the largest pore size of 200 nm achieved a higher pure water flux than the 30 nm membranes, but there were clear differences among the UF membranes. A closer look at tubular UF membranes 7C_s and 7C_r revealed differences in the thickness and number of separation layers. The thickness of the layers has a direct impact on water permeability and is the decisive factor for the observed differences [
28,
29,
30]. Reducing the number and thickness of layers can increase the surface roughness [
31]. This interaction is due to the manufacturing process of ceramic membranes, which are often produced in a dip-coating process in which intermediate layers and a final separating layer are coated to the support material [
32,
33]. The support material provides the mechanical stability, while the intermediate layers reduce the pore size and creates a relatively homogeneous surface. The final separation layer consists of the smallest pores and determines the separation characteristics of the membrane [
32]. The relationship between the number of layers, which increases the cost, and the surface roughness must be taken into account in the manufacturing process. The actual number of layers also depends on the production itself, the technique used is confidential information of the respective manufacturer. Furthermore, differences in water permeability caused by surface roughness have been studied for UF and nanofiltration (NF) membranes. For ceramic UF membranes, roughness was shown not to influence the pure water permeability [
11], whereas studies of polysulfone NF membranes showed that roughness had a positive effect on the pure water permeability. Rougher surfaces feature valleys with lower membrane resistance, and most of the mass flow through the membranes passes through these valleys [
34]. In our case, the membrane with the rougher surface also featured the thinner outer separating layer. Two different effects would therefore occur in the valleys: thinning of the outer separation layer, which decreases membrane resistance, and general thinning of the membrane, further decreasing membrane resistance.
The wall shear stress was used to compare the performance of different membrane pore sizes and geometries. This parameter has already been used successfully to compare ceramic hollow-fiber and tubular membranes with different inner channel diameters [
35]. During the filtration of fermentation broth under comparable process conditions, all three tubular membranes (30 nm and 200 nm) reached a comparable J
ss of 27–33 L m
−2 h
−1 regardless of the pore size, suggesting that steady-state flux was strictly limited by fouling effects occurring under the prevailing hydrodynamic conditions. Similar results were observed using BSA as a feed solution (CFV = 9.5 cm s
−1) with identically constructed flat-sheet polymeric reverse osmosis (RO), NF, and UF membranes [
36]. Despite differences in J
0 reflecting the diverse membrane properties, all membranes reached an almost identical J
ss probably because flux is primarily influenced by interactions between the fouled membrane and the foulant [
36].
Increasing the wall shear stress reduces the formation of a fouling layer on the membrane surface and thus increases flux [
37,
38]. In filtrations with spent sulphite liquor using 8 nm hollow fiber membranes, the effect of increasing flux by increasing wall shear stress was investigated. The authors showed an increase in flux by a factor of ~5 with increasing wall shear stress from 6–130 Pa and consequentially a reduction in fouling [
35]. In our filtrations, the wall shear stress was ~4.5-fold higher on the hollow-fiber membrane than the tubular membranes due to the smaller inner channel diameter and higher CFV, suggesting that the higher J
ss was positively influenced by enhanced particle back-transport. Taking the reversible and irreversible resistance into consideration, the hollow-fiber UF membrane (with the highest J
ss ) also showed the highest irreversible resistance (66%). However, the MF membrane 1C_MF with the largest pore size of 200 nm showed a considerably lower irreversible resistance of 2%. A high proportion of reversible fouling, as we observed for membrane 1C_MF, indicates the formation of external deposits of material that can be removed easily by rinsing or filtration breaks [
14]. Higher wall shear stress reduces the deposition of external material on the hollow-fiber membrane, thus leading to other fouling effects such as internal pore clogging and thus a higher proportion of irreversible resistance. The higher shear stress therefore minimized reversible fouling without influencing the irreversible fouling [
39]. The MF membrane experienced substantially less irreversible resistance compared to the other membranes because the yeast cells reduce membrane fouling caused by proteins. A combination of BSA and
S. cerevisiae limits the fouling of internal membrane structures compared to BSA alone because the yeast cells form a layer that acts like a pre-filter [
40,
41]. Our protein concentration data revealed little to no protein retention. Further experiments are required to study simultaneous biomass separation and the retention or transmission of specific proteins.
The filtration of medium-free cell suspensions using membranes 7C_s and 7C_r at a CFV of 0.8 m s
−1 resulted in a comparable J
ss , indicating that surface roughness does not affect the J
ss in the range studied. J
0 was higher for cell suspensions than fermentation broth, but the J
ss was ~10 L m
−2 h
−1 lower, and the time to reach J
ss was longer. This indicates that other fouling mechanisms are responsible, ruling out the direct comparison and summation of resistances, as per the RIS model.
Figure 7b shows an irreversible resistance of 15% for membranes 7C_s and 7C_r after the filtration of medium-free cell suspensions, showing that differences in irreversible resistance arising from the filtration of fermentation broth must depend on components in the supernatant. Furthermore, although we observed differences in biomass in the retentate during filtration (
Figure 7a), these differences did not affect the flux. This probably reflects a tradeoff between the thickness of the top layer of cells and the thickness of the separation layer. The smaller number of cells in the retentate of membrane 7C_r should ensure that more cell deposition results in greater resistance compared to 7C_s. However, this is offset by the thinner separating layer of membrane 7C_r, thus resulting in a similar J
ss. Given that almost all the biomass could be recovered after filtration by rinsing the membrane, the biomass deposited on the membranes was not a dense cake layer. Indeed, additional filtrations at a CFV of 1.1 m s
−1 revealed clear differences in steady state fluxes. The J
ss increased from 24 L m
−2 h
−1 at 0.8 ms
−1 to 40 L m
−2 h
−1 at 1.1 m s
−1 for the smooth surface of membrane 7C_s, but the CFV had a negligible effect on the rougher surface of 7C_r (J
ss = 20 L m
−2 h
−1 at 0.8 m s
−1 and 17 L m
−2 h
−1 at 1.1 m s
−1). Increasing the CFV reduces superficial membrane fouling and thus increases flux during the filtration of biological suspensions [
42]. In addition, smoother membrane surfaces can promote flux by reducing susceptibility to fouling [
11,
43,
44]. The effects we observed at a CFV of 1.1 m s
−1 indicate that only the smooth membrane has this effect, although the biomass remaining in the retentate increased to 90–100% in both cases. This provides evidence that less foreign material is deposited on the membrane surface, thus limiting the resistance caused by cells. The CFV-dependent behavior of the two membranes should therefore be investigated in more detail.
Comparing the filtration data for the fermentation broth (cells plus medium) and the cell-free medium revealed further interesting phenomena. The J0 of the tubular UF membrane 7C_s and that of 7C_r were comparable, regardless of the feed solution, but a higher Jss was achieved when the resistance caused by yeast cells was absent. We propose that the initial drop in flux during filtrations with cell-free medium is accompanied by the formation of a gel layer, hence the additional presence of cells in the fermentation broth only slightly increases the resistance. A gel layer is not formed during the filtration of cell suspensions without medium, but the cells form a loose filter cake that builds up until the Jss is reached, thus causing more resistance than the combination of a gel layer and cells. Based on these observed resistances, a theoretical resistance can be attributed to the yeast cells even though this cannot be determined experimentally. For the 30 nm membranes, we calculated yeast cell resistances of 1.53 × 1012 m−1 for 7C_s and 2.44 × 1012 m−1 for 7C_r. Although the mass fluxes differ only slightly, the resistances indicate that the smooth membrane 7C_s tends to be less exposed to fouling by the cells, which is likely to reflect the surface roughness.
The 1D profiles and topographic images of 7C_r revealed an irregular surface featuring peaks and valleys. Single line measurements revealed R
a values of ≥5 µm, which is in the same size range as
K. lactis cells and thus increases the probability of particle deposition in the valleys of the asperities of the membrane surface. The deposition of particles in valleys leading to more fouling has already been described as a possible cause for the reduction of permeate flows [
45,
46]. The fraction of irreversible resistance increased slightly during the filtration of cell-free medium compared to fermentation broth, probably because (as discussed above) the yeast cells may act as a pre-filter. The greater irreversible resistance of the 7C_s membrane, which has a smoother surface as well as more numerous and thicker separation layers compared to the 7C_r membrane, was also evident in the filtration of cell-free medium. Medium components that penetrate the membrane pores must therefore be responsible. The difference in the irreversible fraction could be due to surface roughness or the composition of the separating layers, and this can be investigated by SEM. We found that organic material was present at the highest density in the first layer. For the smoother membrane with the thicker first layer, a large amount of densely packed organic material accumulated in the pores, and such material was also present in the second layer, albeit at a lower density. Based on our results, we therefore assume that the thickness of the first layer (and thus the larger surface area in this layer) is responsible for the formation of the higher irreversible fraction due to interfacial interactions between product components and the membrane. Similarly, for the filtration of skimmed milk using membranes that differed only in the thickness of the outer separation layer, the membranes with the thinner separation layer were characterized by lower levels of irreversible resistance caused by the feed [
30].
The membrane properties responsible for fouling can be determined more precisely by investigating additional properties such as the zeta potential of the membrane and feed, membrane chemistry, pore size distribution, porosity, and contact angle in addition to the surface roughness and composition of the separation layers discussed herein [
47].