Protein recovery by selective separation using ceramic membranes

The world's fish and fishery production is growing steadily with a crucial contribution from aquaculture, which continues to expand at an annual rate of 8.8 percent (FAO 2012). As one of the major seafood exporters worldwide, 55% of Vietnam's total fish production originates from aquaculture activities, of which the key species, Pangasius and shrimp, reached a total amount of 1.2 million and 488,000 tons, respectively, in 2012 (Vietfish International 2014). The growth of aquaculture, and especially of the shrimp industry, provides a stable raw material source for the exploitation of chitin and its derivatives. Approximately 40–50%-wt. of raw shrimp is solid waste (Ngoan et al. 2000, Lertsutthiwong et al. 2002). Processing of shrimp residues helps countries producing shellfish to dispose of bio-waste in an ecological and economical manner.

For chitin recovery from shrimp waste on an industrial scale, sodium hydroxide and hydrochloric acid are widely employed as reagents (Lertsutthiwong et al. 2002) to remove the associated proteins (thermal alkaline extraction) and minerals (acidic extraction) in the crustacean shells. Consequently, organic and nitrogenous compounds originate mostly from the alkaline partial stream that accumulates in the wastewater. Water is utilized intensively during production, for washing and purification, and wastewater with a high organic load is also an unavoidable issue for such processing companies.

Because the potential recovery of useful proteins or sodium hydroxide solution from industrial processes has attracted attention recently, studies on alkaline wastewater from chitin processing have been performed. Benhabiles et al. (2013) reported on the possibility of recovering proteins with ceramic ultrafiltration membranes (0.5 μm) and reviewed the influences of chitin processing conditions on the properties of the alkaline hydrolyzate. Filtrations were run on lab-scale, at 20 °C and pH 7, with synthetic deproteinized shell wastewater. Zhao & Xia (2009) combined a stainless steel ultrafiltration (cut-off 20 kD) and a polysulfone nanofiltration (NF) (cut-off 150 D) for caustic solution recovery. Feed solution was sampled from a local shrimp waste-processing factory. Trials were carried out at a technical scale for the ultrafiltration at 70–80 °C, trans-membrane pressures (TMP) of 3 bar, and for the NF at 50 °C, TMP of 31 bar. Both investigations applied a very large range of tangential velocities 4–6 m/s. Chemical oxygen demand (COD) and protein concentrations were relatively low: less than 7 g/L and 1 g/L, respectively.

With respect to protein recovery, membrane technology has been extensively investigated as a process-integrated unit for the separation and enrichment of production-related streams or liquid substrates from food processing, e.g. in cheese-making (Butylina et al., 2006; Galanakis et al. 2014), amaranth starch production (Middlewood & Carson 2012), soy bean flour production (Noordman et al. 2003), and in seafood processing industries (Afonso & Borquez 2002, Bourseau et al. 2009). Researchers focused not only on membrane performances and recovery capacity but also on operational conditions for preventing fouling, a common obstacle in nutrient-rich liquid filtration. Ceramic membranes have been proposed as a solution and are being increasingly applied because of their thermal and chemical stability with regard to influent properties.

This study reports on a comparison of three tubular ceramic membranes for protein recovery from alkaline extraction lysate, under consideration of local chitin processing conditions. Operating parameters, permeate flow rate, and recovery efficiency in pilot-scale testing conditions were evaluated to propose suitable cut-offs related to increasing the concentration of recovered proteins.

The current experiments were performed with multi-channel tubular ceramic membranes: a microfiltration (MF) membrane 0.1 micron (Atech GmbH, Germany) and two NF membranes 450 D, <300 D (Inopor GmbH, Germany) with a filtration surface area of 1.68 m² and 1.75 m², respectively. The cut-offs of membranes were chosen on the basis of previous lab-scale experiments (Nguyen et al. 2014). Each membrane tube has a length of 1.2 m and an active layer of alumina (MF) or of zirconia/titania (NF). Table 1 shows the membrane cut-off (manufacturers' information) and measured values of membrane permeabilities under testing condition with softened water: temperature 35 °C, transmembrane pressure 1.5 bar, and cross flow velocity 3 m/s.

The membrane testing unit is characterized by a feed-and-bleed system supported by a pressure pump and a recirculation pump located in front of the membrane modules. This configuration is identical for both MF and NF. The membrane module splits the alkaline stream (feed stream) into a permeate stream and a concentrate stream.

Experimental performances presented in this paper for three ceramic membranes were carried out with the alkaline stream, in a one-step filtration, at a constant temperature of 70 °C. Prior to the experiments, membranes were rinsed with softened water. The pure water permeability was measured at a temperature of 30–35 °C, at a TMP of 1.5 bar (MF) or at different TMPs 1.5–9 bar (NF) and at a cross flow velocity of 3 m/s. After each experiment, chemical cleaning was carried out to recover the original filtration capacity. The pure water permeability was also re-checked before starting a new trial.

The MF and NF membrane can be backwashed at the pressure of 4.5 bar from the permeate side by pneumatic valves with compressed air and the permeate stream left in the manifold pipe (Figure 1). An integrated backwash was applied impulsively at intervals of 15 and 5 min and with a duration of 4 and 2 s during concentrating runs for MF and NF, respectively.

For the determination of optimal operating conditions, the critical flux phenomenon or critical conditions of the process were investigated. Filtrations were run without backwash in a closed loop. Both concentrate and permeate streams were returned to the buffer tank, to keep the feed concentration constant. Table 2 shows the process with various operating adjustments of pressures and cross flow velocities applied for the tested MF and NF membranes. The process was carried out step-wise. The upward flux value was recorded every 5 s during 10 to 15 min in each pressure step, after the operating conditions had been adjusted.

For the targeted recovery of proteins, filtrations were set in a so-called concentrating mode. This means that the concentrate stream is routed back to the buffer tank, thus generating a concentration increase, while the permeate stream is collected in a storage tank outside the plant. The volumetric permeate recovery (VPR) rate (calculated according to Equation (3)) increases with the running time. In these experiments, the transmembrane pressure was set constantly to 1.5 bar (MF), 3 bar (NF 450), and 5 bar (NF <300). A tangential velocity of 3.5 m/s was applied. To investigate the system stability or membrane performance in a long-term filtration at a constant VPR, the liquid in the buffer tank can be regularly discharged, using the discharge pump during the filtration mode.

Membrane retention R_i,M (Equation (1)) refers to the nearly real retention rate at a given point of time during the filtration process and is calculated using the measured permeate concentration of i c_Pi,M and the bleed concentration c_Bi,M in the recirculation loop of the filtration unit at this moment. System retention R_i,B (Equation (2)) is calculated using the unified permeate concentration c_Pi after the end of the filtration and the concentration of the influent at the beginning of the filtration c_Fi,B.

To assess the nitrogen content, both total bound nitrogen (TN_b) and total Kjeldahl nitrogen (TKN) were determined. TN_b was analyzed according to the Koroleff method (spectrometric method with 2.6-Dimethylphenole after digestion with sodium persulfate) and TKN according to the Kjeldahl method (digestion with H₂SO₄, distillation and titration via Foss 2300 Kjeltec). Because the results for TN_b and TKN are almost identical, values of TN_b are chosen to calculate crude protein (CP) concentration by multiplying with a factor of 6.25 (Ngoan et al. 2000).

For process controlling, samples were also additionally checked with Hach Lange Cuvette Tests (Dr. Lange LCK 514, LCK 314 for COD and LCK 238, LCK 338 for TN_b).

Ammonia (NH₄-N) was determined via photo spectroscopy with the indophenol blue method (DIN 38406-5). Sodium hydroxide concentration was measured by titration with HCl 0.5 N. The total concentration of phosphorus (TP) was measured photometrical using cuvette tests LCK 350 (2–20 mg P/L) of Hach Company. Total suspended solids (TSS) and residue on ignition (ROI) were determined according to the standard German method DIN 38409-2 (1987). For TSS analysis, the sample was filtered with a glass fiber filter (Whatman GF/C, undefined pore size) and dried at 105 °C until its mass became constant. The ignition loss of the filtered residues was determined by heating the sample at 550 °C for 60 min in a furnace.

Similar values for TN_b and TKN indicate negligibly low concentrations of nitrite, nitrate, and ammonia, and also indicated that a significantly large amount of organic bound nitrogen, e.g. proteins, was present in the sample.

Although basic process parameters, e.g. alkali concentration, temperature, solid-solvent ratio, and extraction time, were reported to have a critical effect on the protein removal step (Benhabiles et al., 2013), not all factors could be taken into account in the large-scale production of chitin. The quality of the shrimp waste, the proportion of heads and shells in one charge, the recirculation rate of alkaline hydrolyzate in the first extraction step, etc., profoundly influence the production process and the quality of the alkaline discharged stream. Therefore, analyses varied considerably from batch to batch. The min – max range with a large variation in the investigated parameters corresponds to the real and varying conditions at the production site.

A sustainable flux is desirable in the filtration process to reduce the running costs. Field et al. (1995) presented a critical flux concept for MF. This concept stated that, on start-up, there exists a flux below which a decline of flux with time does not occur; above it fouling is observed. This flux is the critical flux and its value depends on the hydrodynamics and probably other variables (Field et al. 1995, p. 267). Apparently, adapting the operating conditions of the system, namely, transmembrane pressure and cross flow velocity, is one of several methods to minimize fouling and improve the flux (van den Berg et al., 1989). With regard to the issue of critical flux, extensive studies have been performed on MF and UF (Chen et al. 1997; Madaeni et al. 1999; Rayess et al. 2011) (Chen et al., 1997; Madaeni et al., 1999; Wu et al., 1999; Rayess et al., 2011; Sim et al., 2014), as well as on NF (Mänttäri & Nyström (2000), Ellouze et al. 2012), especially in terms of measuring methods for critical flux (Espinasse et al. 2002).

Accumulation of suspended particles or soluble molecules, or both, adjacent to the membrane surface limits the liquid flow through the membrane. Protein-rich feed solutions are considered to be a particular challenge for separation. Due to their viscosity, gels may form on the filter's surface. High pressure or low tangential velocity could accelerate cake layer formation on the membrane surface. High tangential velocity enhances the mass transport through the membrane but is associated with increasing energy consumption. Therefore, optimal hydrodynamics have to be assessed.

The critical flux concept was used to investigate the effect of hydrodynamics on membrane performances. The objective of this study was to identify the optimal pressure and optimal cross flow velocity for membranes MF 0.1 μm, NF 450 and NF <300 in order to filtrate the alkaline process stream steadily. The determination of critical flux was carried out by following the pressure-step measuring method of Mänttäri & Nyström (2000) and Ellouze et al. (2012).

According to Bacchin et al. (2006), the flux performance below the pure water line indicates a very early fouling at the initial stage of the filtration process. At the starting point of this test, the flux-line already deviated from the pure water line and presented itself non-linearly (see Figure 5(b) and 5(c)). This performance is defined as the ‘weak form’ of critical flux (Field et al. 1995) and suits the complexity of treated feed solution properly.

The diagrams 5b and 5c show a similar flux-TMP relationship between the two tested NF membranes. Both exhibited a great flux deviation, especially NF 450, compared to the pure water flux value at the beginning (TMP = 1.5 bar). The permeation of the caustic solution varied from 50 to 158 L/(m²·h) (NF 450) and from 7 to 89 L/(m²·h) (NF <300), as a function of TMP. It slopped considerably at a trans-membrane pressure of 5 bar (NF 450) and 7 bar (NF <300), independent of the applied velocities.

In terms of hydrodynamic effects, especially on membrane permeation and retention, TMP = 1.5 bar (for MF), TMP = 3 bar (for NF 450), TMP = 5 bar and v = 3.5 m/s (for both) were sustainable conditions and are recommended for further trials.

Both NF membranes appeared to have a similar trend, with nearly identical initial permeabilities of approx. J_P = 20 L/(m²·h·bar), whereas the MF started at a higher value of J_P = 48 L/(m²·h·bar). The MF permeate flux sloped continuously with filtration time. The cross flow velocity of 3.5 m/s may have succeeded in reducing cake layer formation but failed to limit the interactions of solid matter and colloids in open membrane channels. The presence of particles, macromolecules, and dissolved substances that cover the membrane surface might induce pore blockage and reduce the liquid flow through the membrane surface. A nearly identical flow performance observed for stainless steel UF 20 kD was reported by Zhao & Xia (2009). Within 15 hours, a VPR of 80% was reached, at which the UF permeability decreased by more than 50% of its initial value.

TN_b was analysed in the concentrate stream to assess the protein retention efficiency. With gradual reduction of the concentrate volume, CP accumulated in concentrate stream from an initial value of approx. 4% to 6.7, 13.7, and 15.8%-wt. by single-step filtration with MF 0.1, NF 450, and NF <300, respectively (Figures 8(b)–8(d) and 9). Whereas MF retains only particulate proteinaceous matter, the increase of CP concentration from ca. 60 to 156 g/L was obtained using only NF membranes. Approximately 80% of the feed volume was recovered as a clear, yellowish permeate with 1%-wt. NaOH. The hot caustic solution is suitable for reuse in the production line. Close to 4% of dissolved nitrogen compounds, e.g. peptides, amino acids, and molecules smaller than 300 D still remain in the permeate stream (Figure 9).

This paper reports on the potential recovery of proteins and caustic solution from the partial alkaline stream of a chitin manufacturer in a tropical region. Three ceramic multi-channel membranes with different molecular weight cut-offs were investigated on a pilot-scale for optimal operating parameters, filtration performance, and protein recovery efficiency. The process stream was identified to be rich in nitrogen, organic substances, and suspended solids. For a sustainable flux, a trans-membrane pressure lower than 5 or 7 bar was proposed as the cut-off for 450 D and <300 D, respectively. A high tangential flow velocity of 3.5 m/s should be applied. Approximately 6.7–13.7–15.6%-wt. of CP can be accumulated in the concentrate with the corresponding MF 0.1, NF 450 and NF <300 membranes. Protein-rich concentrate is an alternative protein source for animal feeding. Furthermore, reuse of sodium hydroxide solution in the chitin production line helps to reduce chemical and heating costs and limits the discharge load to the connected wastewater treatment plant. This concept addresses a sustainable solution for the chitin industry and contributes to maintaining the eco-friendliness of chitin-based products. Characteristics and reuse possibilities of the concentrate stream – a crucial issue – will be the focus of a subsequent publication.

The authors thank the Federal Ministry of Education and Research (BMBF) for funding the research project AKIZ ‘Integrated wastewater concept for industrials zones’ – subproject 4 ‘resource recovery with membrane filtration’ (research grant 02WA1066).