Chapter 4 Ultrafiltration M. J. LEWIS, Department of Food Science and Technology, The University of Reading, RG6 6AP. 4.1 INTRODUCTION Ultrafiltration offers the opportunity to concentrate large molecular weight components without the application of heat or a change of phase. Such components are rejected by the membrane, whereas the permeate produced will contain the low molecular weight components present in the food, at a concentration similar to that in the feed. This results in an increase in their concentration both on a wet weight and dry weight basis in the solution. It is a pressure-activated process, with pressures in the range of 1-15 bar; these pressures are considerably lower than those used in reverse osmosis. For many heat labile macromolecules, e.g. proteins and starches, concentration by UF at ambient temperature will minimise heat-induced reactions which may adversely influence their functional behaviour in foods. Some important functional properties are solubility, foaming capacity, gelation, emulsification capacity, fat and water binding properties. These are discussed in more detail in Section 4.5. In the case of enzymes or pharmaceutical agents, their biological activity needs to be conserved. It also affords the opportunity to separate small molecular weight components from complex mixtures, containing components with a wide range of molecular weights. There have also been investigations into using UF for protein fractionation, but this is not straightforward due to the diffuse nature of the membranes and their selectivity. UF is also very useful for recovering valuable components from food processing waste streams and fermentation broths. Probably the greatest impetus has come from the dairy industry and dairying applications. However, in all applications, flux decline due to concentration polarisation and fouling are probably the two most important practical aspects. 98 M.J.Lewis 4.2 PROCESSING CHARACTERISTICS This section will deal with some of the important processing parameters encountered in ultrafiltration. There are various factors which will influence the outcome of the process, such as the concentration factor and rejection. See Section 3.3. The extent of the concentration is defined by the concentration factor cf), defined as VF/Vc (see eq. (3.5)). Usually the permeate is the biggest fraction by volume. Milk for cheese making is concentrated by UF fivefold, whereas cheese whey is concentrated twentyfold for the production of protein concentrates. Sometimes the resulting permeates are further concentrated by reverse osmosis. 4.2.1 Rejection or retention factors The rejection or retention factor (R) of any component is defined as R = (CF - cp )/cF (4.1) where cF is the concentration of component in the feed and cp is the concentration in the permeate. The rejection is determined experimentally for each component in the feed, by sampling the feed and permeate at the same time and analysing that component. It is very important and will influence the extent (quality) of the separation achievable. Rejection values normally range between 0 and 1; sometimes they are expressed as percentages (0 to 100%). when cp =O; when cp = cF R = 1; all the component is retained in the feed R = 0; the component is freely permeating. In ultrafiltration experiments, some workers have measured negative rejection, Le. cp > cF, particularly for minerals. It is not immediately obvious why this should have occurred. Possible explanations for this are higher concentrations at the membrane surface than in the bulk, due to concentration polarisation. However, this is unlikely to be the case for freely permeating species. Another explanation is the basis on which concen- tration is measured (Glover, 1985). This may arise when there is substantial fat in the feed which is rejected by the material. It is suggested that concentrations be expressed in the aqueous portion. A third explanation lies in the Donnan effect; Donnan predicted and later demonstrated that concentration of electrolyte in the solutions on either side of a dialysis membrane were unequal when the colloid on one side was electrically charged (see later). For example, at low pH values, where proteins are likely to be positively charged, this could lead to higher concentrations of cations in the permeate. Rejection characteristics can readily be determined for different substances using different membranes. This is one practical way of selecting the most appropriate membrane for a particular application. Rejection values may also be influenced by operating conditions. An ‘ideal’ ultrafiltration membrane would have a rejection value of 1.0 for high molecular weight components and zero for low molecular weight components. However, typical values observed for real membranes are between 0.9 and 1.0 for high molecular weights and between 0 and 0.1 for low molecular weight components. Values for Ultrafiltration 99 minerals often are usually in the region of 0.1, but may be as high as 0.5, if the mineral binds to macromolecules. It is important to appreciate that any component with a rejection value greater than 0 will increase in concentration during the course of an ultrafiltration process. Rejection values can be used to check the integrity and performance of a membrane. Some values for components in dairy processing are given in Table 4.1. Note the relatively high values for minerals, which suggests some binding to the proteins, particularly for calcium and magnesium. Membrane manufacturers some- times present performance data in terms of rejection values of a range of components of different molecular weights (see Table 4.2). This will give some guidelines in terms of selection. However, very rarely are those components selected that one is interested in. An alternative form of representation widely used is the molecular weight cut-off value. Table 4.1. Rejection characteristics obtained during ultra- filtration of dairy products Product Proteina Lactose Ash Sweet whey 0.85-1.0 0-0. 2 0-0.5 Acid whey 0.85-1.0 0-0.2 0-0.5 Skim milk 0.965-1 .O 0-0.2 0-0.5 Whole milk 0.965-0.999 0-0.03 0-0.1 a Taken from Lewis (1982). Based on Kjeldahl nitrogen x 6.38. Table 4.2. Some cited rejection characteristics for different components MW 3000 10 000 30 000 100 000 - - - Insulin 6 000 >0.98 Cytochrome C 12 400 >0.98 0.85 0.45 - a-Ch ymotrypsinogen 24 500 >0.98 0.95 0.75 0.20 Albumin 67 000 >0.98 >0.98 0.95 0.30 Adapted from data from Amicon (1992). The molecular weight cut-off values for UF membranes range between about 2000 and 300 000. At values of about 2000, it overlaps with nanofiltration or ‘loose reverse osmosis’, whereas at 30 000 it overlaps with microfiltration. Generally the applied pressure required will decrease with increasing cut-off value and pressures in the range 1-15 bar are used. It is implied that a membrane with a molecular weight cut-off of 5000, would reject all components with that molecular weight value or higher (R = 1) and allow components below that molecular weight to permeate freely. Often dextrins have been used for esti- mating molecular weight cut-off, but these are linear molecules. However, due to the 1.0 c 0 8 0.5 '6 a: .- c 0 - - -4' I Ultrafiltration 101 4.2.2 Yield Ultrafiltration is now being used to concentrate and recover some very valuable com- pounds. The yield or recovery of a component is a very important variable, as it will strongly influence the economics of the process. The yield of a component is defined as the fraction of that component, originally present in the feed, which is retained in the concentrate. For recovery of components it is important to have a high yield. However, when washing out components, such as toxins, the yield should be low. For a batch process, it can be shown that the yield of any component depends upon the concentration factor and rejection. Concentration factor cf) = vF/vC (4.2) and the yield (Y) is given by mass component in final concentrate - VCCC Y= -- mass substance in feed VFCF (4.3) where V, and VF are the volumes of feed and concentrate and cc and CF are the concen- trations in the concentrate and feed. If we now consider a batch concentration process depicted by Fig. 4.2, where permeate is removed and the retentate is recycled: At any instance let the volume of the concentrate = V and the concentration of the component of interest = c in" f" Permeate CP Fig. 4.2. Batch concentration process for yield calculation. Let the removal of a small volume of permeate (dV), result in a change of concentration (dc) * A mass balance on the component will give the following equation: VC = (V-dV)(c-dc) + cpdV (feed) (concentrate) (permeate) Rejection (R) = (c - cp )/c 102 M. J. Lewis Thus cp=c(l -R) Eliminating cp gives VC= (V-dV) (c-dc)+c(l-R)dV - Vdc = cR dV (Note: dVdc is assumed to be negligible.) dV cc dc 1 - -- I v -IcF -zi Integration between the final and initial conditions gives: In (vF/vC)=l/R In (cC/cF) (4.4) vF/Vc = f = (cC/cF)l'R If In (VF/Vc) is plotted against In (cc/cF), the gradient is 1/R. (4.5) From eqs. (4.2) and (4.3), it can be shown that CC/CF = yf Substitution into eq. (4.5) gives vF/vC =(fY)'lR Therefore f = (fY)'IR. This simplifies to Y = f '-'. Therefore the yield y= fR-1 (4.6) However, this equation applies only if the rejection remains constant. Nevertheless, it is extremely useful, as it gives an insight into the features of the separation process. Let us consider the two extreme values of rejection: If R = 1, then yield = 1; all the material is recovered in the concentrate. If R = 0, then yield = l/J in this case the yield is determined by the concentration factor. As the concentration factor is finite (typically 2-20), the yield can never be zero; i.e. it is not possible to remove all of a component from a feed by ultrafiltration alone. Diafiltration may be more useful in helping to achieve this objective (see Section 4.4). However, for most components being concentrated, the rejection values are close to 1.0, typically 0.9-1.0, whereas for those being removed the values would be between 0 and 0.1, Table 4.3 shows a range of yield values for some different concentration factors. One interesting point is that losses can be quite high, even though the rejection value appears good; e.g. for R = 0.95 and a concentration factor of 20, the yield is 0.86. Therefore 14% Ultrafiltration 103 Table 4.3. Yield values for different concentration factors and rejections Concentration factor Rejection 0 0.1 0.2 0.5 0.9 0.95 1.00 2 0.50 0.54 0.57 0.71 0.93 0.97 1.0 5 0.20 0.24 0.28 0.45 0.85 0.92 1.0 10 0.10 0.13 0.16 0.32 0.79 0.89 1.0 20 0.05 0.07 0.09 0.22 0.74 0.86 1.0 50 0.02 0.03 0.04 0.14 0.68 0.82 1.0 of the component is lost in the permeate. Yield values are also sometimes quoted as percentages. However, this equation gives the maximum yield, which would be for a batch process. The yield is likely to be lower for a continuous single or multistage process, simply because steady state is achieved at higher levels of concentration. For such a process the yield is given by 1 Y= f - R(f -1) The concentration of a component in the final resulting concentrate (CC) can be calculated from the following equation: CC = CF Yf (4.7) cc=cF fR (4.8) or However, there is some evidence that rejection does not remain constant. During a batch ultrafiltration experiment the rejection of most components rises, as has been observed on many occasions (see Fig. 4.3). 4.2.3 Average rejection In situations where the rejection does change significantly, an alternative evaluation procedure is to measure the yield for the process, and then to work backwards to calculate the rejection value, which would have given rise to that yield. This rejection value is termed the average rejection value (Rav) (4.9) 1 y - vccc - cc VFCF CF f If this expression for yield is equated with that from eq. (4.6): 0.4 0.2 0 - - IIIIIIIII Ultrafiltration 105 for most food systems. Therefore it is very important to measure rejection data under the prevailing operating conditions. Lewis (1982) has compiled rejection data for different systems. It was not always clear whether rejection data for proteins was based upon crude protein or true protein. Ultra- filtration could be useful for removing non-protein nitrogen. There also appeared to be some confusion between the terms rejection and yield in some of the earlier reports. Table 4.1 shows some rejection data for some dairy products, reported by the Interna- tional Dairy Federation (1979). Figure 4.3 shows rejection data taken during the batch ultrafiltration process during concentration of rapeseed meal, for crude protein, total solids and glucosinolates. For all components, there is an increase in rejection as concentration proceeds, with the increase being most marked between concentration factors of 1 and 2. Many investigators have reported similar increases in rejection as concentration proceeds. Table 4.4 shows some data for the average rejection data for proteins and glucosinolate, extracted at different pH values, determined by the method above. Yield values are also presented. In such complex systems the performance is also strongly affected by pH (see Sections 4.3.2 and 4.5.2). Table 4.4. Average rejection (Rav) and yield values for glucosinolates and crude protein during batch ultra- filtration processes at different pH values. PH Glucosinolate Crude protein 2.5 0.50 (0.45) 0.97 (0.95) 3.5 0.39 (0.38) 0.93 (0.89) 7.0 0.28 (0.31) 0.81 (0.74) 9.0 0.36 (0.36) 0.95 (0.92) 11.0 0.44 (0.41) 0.85 (0.92) Yield values in brackets. Glucosinolates are expressed as isothiocyanates. Therefore, rejection values are very impofiant as they influence the nature of the separation obtained, as well as the yield (or loss) of components. These aspects assume greater importance as the value of the product increases. Changes in rejection during a process could also be indicative of some important changes taking place at the surface of the membrane. The effects of pressure and temperature on rejection, as predicted by some of the models, are discussed in Chapter 3. Some practical problems associated with UF of proteins, such as adsorption and pH effects, are described by Sirkar and Prasad (1987). 4.3 PEKFORMANCE OF ULTRAFILTRATION SYSTEMS Permeate flux In UF process applications, the two most important parameters are the membrane 106 M. J. Lewis rejection (see also Chapter 3) and the flow rate of permeate or permeate flux, hereafter abbreviated to 'flux'. The flux will probably be measured in gallons/min or litres/hour, but it is usually presented in terms of volume per unit time per unit area (1 m-2 h-l). Expressed this way it allows a ready comparison of the performance of different mem- brane configurations with different surface areas. Flux values may be as low as 5 or as high as 450 1 m-* h-'. The flux is one of the major factors influencing the viability of many processes. UF processes have been subject to a number of modelling processes, in an attempt to predict flux rates and rejection values from the physical properties of the solution, the membrane characteristics and the hydrodynamics of the flow situation, in order to opti- mise the performance of the process. 4.3.1 Transport phenomena and concentration polarisation Ultrafiltration is usually regarded as a sieving process and in this sense the mechanisms are simpler than for RO. However, it is important to remember that for pressure-driven membrane processes, the separation takes place not in the bulk of solution, but in a very small region close to the membrane, known as the boundary layer, as well as over the membrane itself. This gives rise to the phenomenon of concentration polarisation over the boundary layer. (Note that in streamline flow the whole of the fluid will behave as a boundary layer.) Concentration polarisation occurs whenever a component is rejected by the membrane. As a result, there is an increase in the concentration of that component at the membrane surface, and a concentration gradient over the boundary layer. This increase in concentra- tion offers a very significant additional resistance, and for macromolecules may also give rise to the formation of a gelled or fouling layer on the surface of the membrane (see Fig. 3.4). It is interesting to note that the boundary layer does not establish itself immediately at the point where the fluid first contacts the membrane. Rather it takes some distance for it to be fully established. This distance taken for it to be fully established has been defined as the entry length, and the process of establishment is illustrated for a tubular membrane in Fig. 4.4. Howell et al. (1990) have analysed flux conditions over the entry length and have concluded that the flux and wall concentrations change quite consider- ably over the developing boundary layer, although changes were less marked for a fouled membrane. There would also be less likelihood of operating in the pressure-independent Permeate Membrane Feed inlet --+- Retentate Boundary layer Membrane Permeate Fig. 4 4. Development of the concentration polansation or boundary layer. Ultrafiltration 107 region in the entry length, so there could be an improvement to the flux by use of higher pressures. They reported that the entrance length may be greater than 1 m, but that most of the benefits to flux which could occur using higher pressures would be over the first 20 cm. One much used model considers a number of resistances in series. Therefore, during the transfer of components from the bulk of the solution to the permeate, the main resistances are due to the membrane (Rm), the fouling layer (Rf) and the polarisation layer (Rp). Therefore the flux can be expressed as J = AF'/p(R, + Rf + RP) (4.12) where p is the viscosity of the solvent. The pressure term may be modified to (AP - Az), to account for differences in osmotic pressure, but in most UF applications, the osmotic pressure differences (An) are negligible. This type of model is sometimes known as the 'resistance in series model'. In practical terms, the effects of concentration polarisation can easily be seen, as there is a marked reduction in flux, when water is replaced by the solution to be ultrafiltered, using either a dynamic or static start (Fig. 4.5). For a new membrane, flux data is determined using water, before use, as this provides an indication of the condition of the original membrane and its resistance (R,), or a membrane after cleaning. Membrane-cleaning protocols are designed to restore the water flux back to its original value. As soon as the water is replaced by the fluid, the flux rate will fall by a factor of 2-10 times, in a very short period of time, usually less than one minute. Thus the equilibrium described is achieved in a relatively short period of time. As concentration proceeds, the flux will further decline, due to a combination of an increase in the viscosity (total solids) and the process of fouling. However, it is not easy to separate the effects of polarisation, fouling and concentration increases. Therefore, experiments to observe the effects of fouling are usually performed at constant composition, by returning the permeate to the feed tank. Also, it is not straightforward to assess the individual contributions of fouling and concentration polarisation to the flux decline during the initial transition from water to product. Some fouling is evident with most systems and is assessed by the decline in flux 1 3 E 2 3 Time Fig. 4.5. Flux decline due to concentration polarisation and fouling: (1) water flux; (2) moderate flux decline; (3) rapid flux decline. 108 M. J. Lewis rate with time, usually for a period of 30-120 min, at constant composition. Fouling will be discussed in greater detail later. In the absence of fouling, there are two main transport steps: (1) through the boundary (polarisation layer) and (2) through the membrane (Aimar, 1987). The first depends upon the hydrodynamics, on flow rate of solvent to the membrane (permeate flux), fluid composition and transport properties. The second depends upon the applied pressure and the properties of the membrane, namely its average pore size, distribution of pore size and its chemical properties. One of these transport steps is likely to be the rate limiting process. In general terms, the effects of operating parameters are shown in Fig 4.6. It should be noted that results are obtained for a constant composition. As the concentration of the feed increases, the flux will further decline. It can be seen that there is a pressure- dependent region (AB) and a pressure-independent region. However as the flux rate increases, the rejected materials will increase in concentration and concentration polarisation becomes limiting. In this region flux rates can be increased by higher flow rates and operating temperatures. Therefore this resistance approach provides an explanation for the observed occurrence of pressure-dependent and pressure-independent regimes. At low pressures and low flux rates, the membrane offers the controlling resistance. This would suggest that ultrafiltration processes are best operated at pressures corresponding to the initial onset of the pressure-independent region. Use of higher pressures would only be wasteful of energy. There is some evidence that this could also give rise to a decrease in the flux, due to compaction of the membrane or fouling layer. In the pressure-independent region, the following analysis has been performed to model the flux performance, based upon a material balance at the membrane surface. This is known as the film theory model (see Fig. 4.7). A dynamic equilibrium is established, where the convective flow of the component to the membrane surface equals the flow of material away from the surface, either in the permeate or back into the bulk of the solution by diffusion, due to the concentration gradient established. This is expressed as X x LL LL in viscosity In (conc.) (b) Pressure (4 Fig. 4.6. (a) Effects of operating pressure and flow rate on flux; (b) flux decline measured against concentration factor. II 1: ~I ~1 11 110 M. J. Lewis Table 4.5. Gel concentrations for different proteins Feed Gel concentration cg Skim milk 20-25% protein Full cream milk (3.5% fat) 9-1 1 % protein Acid whey 30% protein Sweet whey 20-28% protein Gelatin 22-30% protein Egg white 40% protein Defatted soy extract 20-25% protein Taken from Cheryan (1989). these values and those determined for the proteins by more conventional methods, throwing some doubt on the validity of this type of analysis (Aimar, 1987). A further explanation for the discrepancies could be that the rejection of the components is not 1.0, making eq. (4.15) not directly applicable. Nevertheless these equations help explain in qualitative terms the observations that in the pressure-independent region, the flux can be increased by increasing the mass transfer coefficient. Because of this, considerable attention has been paid to the determination of this mass transfer coefficient. There are various qualitative (empirical) relationships in the literature, which correlate mass transfer coefficient to physical properties, flow channel dimensions and operating parameters. Cheryan (1989) concluded that many of these are not very satisfactory. Dimensional analysis has also been frequently used: Sh =A Rea Scb where Sh is the Sherwood number = (kd/D), Re is the Reynolds number = (vdp/p) and Sc is the Schmidt number = (p/pD), where d is the tube diameter and A, a and b are constants. For other flow situations the hydraulic mean diameter can be used, which equals 4 (cross-sectional area/wetted perimeter). A much used form of the equation for turbulent flow is Sh = 0.023 Reo.8 Sc0.33 (4.16) Cheryan (1986) summanses some of the constants for different flow geometries and feed materials. Under turbulent flow conditions, the constant a ranged between 0.5 and 1.1, with 0.8 being a typical value. The value was 0.3 for laminar flow. Again dimensional analysis predicts that the flux rate can be increased by increasing the Reynolds number and by inducing turbulence. However, it has been noted that high flux rates have been observed at high shear rates under streamline flow conditions. More recently, Colman and Mitchell (1991) have described how pulsed flow and baffles have increased flux rates by a factor of 3, at a Reynolds number of 100. This was stated as giving a mass transfer value equivalent to a steady Reynolds number of 10 000 in unbaffled channels. Ultrafiltration 11 1 Although dimensional analysis predicts the importance of turbulence, the values predicted using these models are often lower than those measured in practice. Measure- ment often requires an estimate to be made for c,, the concentration at the membrane surface. The reasons for these higher experimental values are attributed to factors other than diffusion, causing the transport of rejected materials back into the bulk of the solution. One such explanation for colloidal particles is the ‘tubular pinch effect’, whereby colloidal molecules were observed to move away from the tube wall. Aimar (1987) describes the ultrafiltration of pseudoplastic fluids and reported that in the pressure-independent region, the value of the limiting flux JL was found to be given JL = Auacb (4.17) which fits in well with conventional mass transfer theory. In short, the information derived from modelling gives qualitative data on the ways of improving mass transfer. Some of the drawbacks of modelling for real systems stem from the complexity of real feeds. Most are multicomponent, with many compounds which are totally or partially rejected, with interactions between components. Their rheological behaviour is complex, and there are difficulties of measuring the physical properties of the solutions under conditions found within the membranes and a lack of accurate diffusion data for macro- molecules. There is a further suggestion that osmotic effects are likely to be more important than initially thought for ultrafiltration, because the osmotic pressure difference over the membrane depends upon the concentration at the membrane surface and not on that in the bulk solution. However, those components which contribute most to the osmotic pressure have a low rejection and therefore show little accumulation at the membrane surface. When the membrane system is operating at very low transmembrane pressures, say about 1 atmosphere, the osmotic pressure difference may be significant. 4.3.2 Fouling In most practical applications, fouling of the membrane takes place and this is a major operating problem in ultrafiltration. Fouling material collects on the surface of the mem- brane (and perhaps internally) and gives rise to a steady decline in the permeate flux (see Fig. 4.5). This could be particularly important for continuous processes operating at steady state, where a long-term decline in the flux would be extremely detrimental to the process. It could also give rise to a reduced life for the membrane, due to more stringent cleaning regimes being needed to remove the fouling. Fouling is almost impossible to avoid. Removal of colloidal and particulate matter is of paramount importance prior to processing and should always be carried out. However, both Fane and Fell (1987) and McGregor (1986) have described situations where fouling was apparent, even with ‘pure’ water. McGregor also reported that prolonged exposure to 200 ppm of sodium hypochlorite caused considerable flux decline. When more complex materials are involved, such as proteins, interactions can occur between the proteins and the membrane material; for example proteins may bind to the membrane by hydrophobic effects, charge transfer such as hydrogen bonding and electrostatic interactions, or through combinations of these. Conditions that minimise the amount of binding to the by 1 12 M. J. Lewis membrane should be useful in reducing fouling. The two characteristics which appear to strongly influence fouling are the physicochemical properties of the membrane and porosity and morphology of the surface. It is also not easy to assess the individual contri- butions made by concentration polarisation and fouling toward flux decline. Wu et al. (1991) suggest that concentration polarisation is responsible for the rapid initial flux decline, which is followed after about ten minutes by a long and gradual flux decline, caused by fouling. Fane and Fell (1987) report that the flux decline due to concentration polarisation is largely reversible and is therefore different in nature to that caused by fouling. Work on measuring the membrane resistance (R,) during these early stages of concentration polarisation showed an increase in its value, suggesting that some fouling was taking place within the pores, as well as on the surface. Other suggestions are that there are three phases in flux decline: (1) due to concentration polarisation, taking place very quickly (seconds); (2) due to initial adsorption of protein onto the surface; and (3) due to further adsorption and an increase in deposit thickness. There have been relatively few attempts to present a mechanistic model for long-term flux decline, because the issues involved are extremely complex. It has been pointed out by Wu et al. (1991) that the adsorption of protein on the membrane alone can change the filtration characteristics, and that the amount and rate of protein adsorption vary with many factors, including the nature of the membrane, type of protein, its concentration, pH and ionic strength of the protein. Any attempts to model the process taking place needs much experimental data and the parameters depend on the conditions used. One such attempt has been described by Suki et al. (1986), which assumes that protein aggregation in the concentrated layer next to the membrane is based on flocculation theory. Using data available for BSA, the flux is calculated as the aggregate layers build up. Qualita- tively, the model predicts increased flux decline for higher feed concentrations. However, it does not allow for differences in the membrane properties. Owing to these difficulties, there are also a number of models (empirical, phenomenological) to describe the fouling process. Unfortunately they do not offer an explanation for the reasons why fouling occurs, although they are useful for predicting flux decline with time. Some of these have been reviewed by Cheryan (1986), Fane and Fell (1987) and Wu et al. (1991). The simplest is based on an exponential relationship between flux and time: J = Joe-bt (4.18) where J = flux at time (t) and Jo = flux at time zero and t is time or J = JoV-b (4.19) where V = volume permeated and b is an index of fouling. One drawback of these equations is that they predict that the flux will eventually decline to zero, which is not the case in practice. Cheryan (1986) has listed seven models based on this exponential decay. Other types of model use the resistance in series approach, where flux decline results from an ageing process in the deposit. Some of these are reviewed by Fane and Fell (1987). More recently, Wu et al. (1991) have developed a semi-empirical model, which is based firstly on postulating that the rate of flux change is Ultrafiltration 113 directly proportional to the flux itself, and secondly using an exponential attenuation term to describe the effect of membrane fouling on the rate of flux decline. The main param- eters in the model are a rate constant related to concentration polarisation (kp) and a rate constant related to membrane fouling, which can easily be determined, due to a linear relationship between ln(1n (J/Ji)) and time where Ji is the final (steady) flux. Data is presented for a whey protein isolate and the model is also used to estimate these param- eters for some other fouling systems. Considerable advances have been made in understanding the factors which affect flux rates and fouling for dairy products, particularly with whey processing. This is considered in more detail as it provides a good illustration of the complex interactions of solutes during an ultrafiltration process. The compositions of sweet and acid wheys are given in Table 4.6. More details are provided by Hayes et al. (1974), Lee and Merson (1976), Glover (1985), and Heng and Glatz (1991). Fouling characteristics were different in sweet whey and acid whey. Initial fluxes are slightly higher for sweet whey than acid whey, which was attributable to its having only about half the amount of calcium. However, fouling has been reported to be just as severe in sweet whey as in acid whey, attributed mainly to calcium phosphate. For sweet whey, the flux can be improved by the following procedures: heating at 6OoC for 30 min prior to ultrafiltration, thereby precipitating the calcium phosphate; demineralisation by ion exchange or electrodialysis (see Chapter 6), the reduction in ionic strength reducing protein aggregation; using cal- cium sequestering agents, such as EDTA and sodium hexametaphosphate; and lowering the pH to about 3, which increases the solubility of calcium phosphate. An alternative procedure is to demineralise the whey, adjust the pH to 4.6 and allow the deposit to settle out. For acid whey, pasteurisation and adjustment of the pH to between 5.2 and 5.9 brings about an improvement in flux, and this is more marked for hydrochloric acid whey compared to sulphuric acid or lactic acid wheys (Hayes et al. 1974). Kuo and Cheryan (1983) found that fouling could be minimised by pH adjustment to between 2 and 3, followed by removal of resulting insoluble particulates. At low transmembrane pressures it was observed that higher flow rates had a beneficial effect on fouling, whereas at higher transmembrane pressures, higher flow rates significantly increased the fouling rate. They suggested that fouling takes place initially on the surface, which is followed by Table 4.6. Composition of sweet whey and acid whey and skim milk (% w/v) Sweet whey Acid whey Skim milk Lactose 4.9 5.1 4.8 Protein 0.5 0.6 3.3 NPN 0.04 0.04 0.04 Fat 0.3 0.1 0.15 Ash 0.6 0.7 0.7 Lactic acid 0.2 tr - PH 5.7-6.4 4.1-4.4 6.6-6.7 114 M. J. Lewis fouling within the pores. Patocka and Jelen (1987) found that all treatments resulting in the elimination of free calcium improved flux. Heng and Glatz (1991) determined that the extent of flux decline increased with the level of calcium and phosphorus deposition during processing. Electron microscopy has been used to observe the fouling deposits. Lee and Merson (1976) showed the complexity of fouling; several types of morphology depending upon the treatments and conditions used. Glover (1985) reported that deposits 1 pm in thickness have been observed, with some of the particulate matter penetrating up to 2pm into the membrane (Nisbett et al., 1981). Of the proteins present, fouling is mainly attributable to p lactoglobulin, which forms sheets over the membrane surface. More advanced analytical techniques now available, such as IR and X-ray photoelectron spectroscopy can be used to identify fouling compounds and residues after different cleaning regimes (Daufin et al., 1992). Some recent work with permeates from different types of milk and whey has shown that some of the low molecular weight components in milk, such as calcium and lactose, also cause fouling to take place (Ramachandra Rao et al., 1994). This is contrary to the evidence of Kessler er al. (1982). Fouling and concentration polarisation behaviour have been compared by McGregor (1986) for a wide range of commercially available membranes, using a number of com- pttn piuiciir CXLIALL~. rrc uemeb ilriu' u5cs a poiari'san'on/&~ulifig ina'ex and a r6uIing index to compare their performance. To summarise, during an ultrafiltration process the flux declines. There is usually a rapid decline when the process fluid replaces the water due to concentration polarisation. Further decline will result from the increase in the bulk viscosity of the feed and the onset of fouling. When the flux reaches a predetermined low value, the membranes will need cleaning in order to restore the flux. Methods and strategies for reducing fouling are discussed by Fane and Fell (1987). 4.3.3 Factors affecting flux Energy iriput The main source of energy is the pumping energy. The power utilisation (w) is related to the pressure (head) developed and the mass flow rate as follows: W = rn'hg (4.20) where nz' = mass flow rate (kg s-'), h = head developed (m) and g = acceleration due to gravity (9.81 m sk2). This energy is largely dissipated within the fluid as heat and will result in a tempera- ture rise. Cooling may be required to prevent this. Factors that improve flux rates usually increase energy consumption. There has been a great deal of experimental investigation into the factors affecting the permeate flux. For a simple system, the average pressure of the product is (P1 + P2)/2, where P, and P2 are the inlet and outlet pressures respectively. In most cases the pressure on the permeate side of the membrane is atmospheric. Therefore, the average pressure driving force over the membrane is given by (PI + P2)/2; Ultrafiltration 115 this is often referred to as the transmembrane pressure. Pressure may also be applied to the permeate side, in order to reduce the initial flux and concentration polarisation during the transition from water to product. The pressure difference (P1 - P2) will depend upon the flow rate and viscosity of the fluid. The processing variables which have been found to be important are as follows: (1) Pressure. An increase in the pressure increases the permeate flux, usually up to a limiting value, above which the permeate rate becomes independent of pressure: for these reasons we talk of a pressure-dependent and pressure-independent region. The pressure-independent region has been linked with the formation of a gel- permeation layer, adjacent to the membrane. Higher pressures are said to cause compaction of the gelled layer near the membrane surface and perhaps the mem- brane itself. Flow rate. Increasing the flow rate increases turbulence and decreases the boundary layer thickness; this increases the permeate rate; the relationships between permeate rate, pressure and flow rate are shown in Fig. 4.6. An increase in the flow rate leads to an increase in the pressure, at which the onset of pressure-independence begins. Higher flow rates give rise to higher frictional losses and higher energy require- ments: for streamline flow conditions: pressure drop = (flow rate)',' for turbulent flow conditions: pressure drop 0~ (flow rate)'.*. Temperature. All investigators have found that increasing the temperature increases the permeate flux; the relationship between permeate flow rate and temperature is almost linear over the relatively narrow range of temperatures investigated. An increase of about 2% per "C rise in temperature is quite normal. Suggestions for the increase in permeate rate are: a change of porosity of the membrane, decrease in viscosity of the feed, higher diffusion rates or increased solubility of the diffusing material in the membrane. In practice, for foodstuffs, chemical and microbiological considerations have a big influence on the operating temperatures used. Temperatures between 5 and 45°C are often avoided, as microorganisms may grow. With non-cellulosic mem- branes, the highest temperatures which cause no loss of activity or functionality are used; often in the temperature range 50 to 60°C. Higher temperatures (70-90°C) are used for cleaning and disinfecting the equipment. As well as the operating conditions, the nature and composition of the feed affect the permeate rate; particularly its chemical nature and its viscosity. More detail will be provided in the applications section. There is some interdependence; for example, temperature affects viscosity and all the latter three factors influence the degree of turbulence. Turbulence. The role of turbulence has been described in an earlier section. The extent of turbulence is also influenced by temperature and composition and is characterised by the Reynolds number (eq. (4.16)). The two extremes are laminar flow conditions, usually combined with high shear rates, through to fully developed turbulent flow. (2) (3) (4) 116 M. J. Lewis Other factors such as pH and osmotic pressure may also influence the permeate flux, as well as membrane configuration (see Chapter 3). 4.4 DIAFILTRATION Introduction An extension of ultrafiltration employs the addition of water at some stage during the concentration process. It should be remembered that during ultrafiltration, the concentration of any component in the retentate can never decrease (unless there is a true negative rejection). At best, for zero rejection, it will remain constant. However the amount (yield) of a component of low rejection value is significantly reduced, as is the dry weight composition, compared to a substance with a much higher rejection. To effect a reduction in concentration, the retentate needs to be diluted with water; such a process is known as diafiltration. The net effect of diafiltration is to wash out more of the lower molecular weight components. The two main modes of operation are discon- tinuous dinfiltration (DDF) and continuous diafiltration (CDF). DDF is where the water is added back to the retentate intermittently. For example a dilute extract may be concentrated by a factor of 10, by normal ultrafiltration. Water could then be added back to the extract, say to restore the original volume. The diluted retentate could then be further processed (diafiltered) to further reduce the volume of the retentate. Further water can be added as required. Extensive diafiltration should eventually reduce the concentration of low molecular weight compounds to very low levels. DDF can be analysed in the same way as a batch ultrafiltration process, with account being taken of dilution effects. Lewis and Finnigan (1989) found DDF useful for removing glucosinolates from rapeseed meal extracts. After a fivefold concentration, followed by dilution back to the starting volume and a further fivefold concentration, the protein content had increased from 45.3 to 83.0% (dwb) and the glucosinolate concentration was reduced from 2.5% to 0.3%. However glucosinolate rejection values were greater than zero and there were considerable losses of protein in the permeate. There are many similar examples of diafiltration processes in the literature. 4.4.1 Washing out at constant volume CDF involves the continuous addition of water, at a rate equal to the rate at which permeate is removed. Normally this would be termed washing out. This is probably the most common form of diafiltration and can be subjected to mathematical analysis (see Fig. 4.8). Let system volume = V, permeate flow rate = P water flow rate = W At steady state W = P Ultrafiltration 117 Water (w) .1.1' Permeate (P) (4 * Wash-in fluid (b) Fig. 4.8. Diafiltration: (a) washing out; (b) washing in. Solute balance: V~C = Vo (C + dc) + Pc~ dt -Vo dc = Pc~ dt For R = 0, c = cp -Vo dc= Pcdt c dc - Vojco 7 =P dt -V, In - = Pt (4.21) Pt v (9 e 1 vo vo In - =-=- where V = cumulative permeate volume. The ratio V/Vo is termed the number of diavolumes removed. Equation (4.21) gives the number of diavolumes required to bring about a certain removal of the freely permeating species (R = 0). Thus, a tenfold reduction in concentration would require the removal of 2.303 diavolumes. When R # 0; cp = c (1 - R) -Vo dc ~(l - R) -- - P dt -Jc -Vo c -=/Pdt dc (1-R) o c '0 In - = Pt - (1 - R) (9 118 M. J. Lewis Therefore In - =(l-R) - (4.22) (9 (J Again, this equation is only applicable if R remains constant. Thus a component with a rejection of 0.1 would require the removal of 2.59 diavol- umes to reduce its concentration ten times. (cf. 2.3 diavolumes for R = 0.) Glover (1985) stated that in milk, the most efficient diafiltration was to concentrate to about 2, followed by continuous diafiltration. If the analysis is done on the permeate, the equation becomes In( co(fp R)) = (1- R)( t) (4.23) This shows how the permeate concentration changes with the number of diavolumes, or with time. All these analyses have ignored the void volume V’, which can be accounted for by an additional term in eq. (4.2 I), as follows: In(c/cg) = V/V~ - v’/v~ Analysis shows that the optimum concentration for the retained species (R = l), to mini- mise the processing time, is given by cg/e, where cg is the gel layer concentration (see Table 4.5). Further analyses for optimising the processing time for diafiltration of soya bean extracts and cheese whey are performed by Ali Asbi and Cheryan (1992). The optimum appeared to be a combined process where the feed is first concentrated by ultrafiltration, followed by continuous diafiltration. Washing-in Washing-in applications can be analysed in a similar way: In - - (4.24) for R = 0 ( c;:c)= ; for R # 0 In ( -) = ( ; ) ( 1 - R) (4.25) where cf = concentration of component in the wash-in fluid. Washing-in procedures have been used in binding studies, for example protein solution is placed In the cell and the binding component is added in the wash-in solution. Analysis of the component in the permeate together with a knowledge of its rejection characteristics and a mass balance will permit the total quantity of the component bound to the protein to be calculated. Further examples and details are provided by Cheryan (1986). U ,/' ,/' ,/' i\ / - p2 ,' x Feed - ,/' ,' ,/' ,,e' 120 M. J. Lewis Table 4.7. Membrane data for protein fractionation Membrane MWCO Rejection (ideal) Rejection (real) A BC A B C X 10 000 0 1 1 0.1 0.99 1.00 Y 100 000 0 01 0.05 0.10 0.99 Table 4.8. Composition of different streams Composition (dry weight basis) Ideal Real A B C A B C Feed 33.3 33.3 33.3 33.3 33.3 33.3 c2 0.9 9.0 90.1 1.3 11.0 87.7 p2 9.1 90.9 0 11.3 86.4 2.3 PI 100 0 0 97.4 2.6 0 Further purification of each stream could be accomplished by more extensive diafiltra- tion. It can be seen that although there is some scope for enrichment of proteins, true fractionation is difficult. The major limitation would appear to be due to the diffuse nature of the membrane and the wide distribution of pore sizes. Cheryan (1986), using a similar type of analysis, came up with a similar set of data and arrived at similar conclusions. An alternative procedure would be to employ continuous diafiltration to wash out each of the proteins in turn, starting with the smallest. McGregor (1986) has examined a wide range of membranes for separating complex mixtures of proteins. Some membranes showed reasonably sharp characteristics in terms of the permeate, i.e. it was possible to produce permeates with very little protein above a certain molecular weight, but with a mixture of proteins in the concentrate. A novel approach to enrichment, in situations where the fraction of interest represents a small proportion of the total protein, has involved depositing inorganic or organic compounds within the matrix of a porous stainless steel tube. These components separate on the basis of charge or size exclusion. Immunoglobulin G has been enriched from 8% to 20% in chccse whey by this methqd (Thomas et ul., 1992). 4.5 ULTRAFILTRATION APPLICATIONS Some applications of iiltrafiltration will now be discussed, which take advantage of the opportunity to concentrate macromolecules or to remove small molecular weight components, at ambient temperature, without changing pH or ionic environment. Ultrafiltration 121 A wide range of membrane materials is available, the most common being cellulose acetate, polyamides, polysulphones and polyethersulphones; each with different flux characteristics, rejection values, and other physicochemical characteristics, such as charge and extent of hydrophobicity. Molecular weight cut-offs range from 2000 to 300 000, with operating temperatures up to 80°C, over the pH range 1 to 14. Ultrafiltration is also very useful for recovering valuable components from food- processing waste stream and fermentation broths. Probably the greatest impetus has come from the dairy industry and dairying applications. However, in all applications, flux decline due to concentration polarisation and fouling are probably the two most important practical aspects. 4.5.1 Dairy applications Milk is chemically complex, containing components of a wide range of molecular weights, such as protein, fat, lactose, minerals and vitamins. It also contains micro- organisms, enzymes and perhaps antibiotics and other contaminants. An idea of the complexity of milk is given by Walstra and Jenness (1984), who list well over 50 components. Whole milk contains about 30-35% protein and about the same amount of fat (dry weight basis). Therefore it is an ideal fluid for membrane separation processes, in order to manipulate its composition, thereby providing a variety of products or improving the stability of a colloidal system. The same applies to skim-milk, standardised milk and some of its by-products such as cheese whey. Skim milk can be concentrated up to seven times and full-cream milk up to about five times (Kosikowski, 1986). Milk can be derived from a number of different species, cows' milk being the most common, with milk from buffalo, goats and sheep being drunk in substantial quantities throughout the world. Milk is either consumed in its liquid form or converted to a wide variety of products. Surplus milk is usually preserved as skim milk powder and butter or anhydrous butterfat. At one time the most valuable component was the fat, with cream products and butter fetching high returns. Skim milk was therefore a by-product of cream and butter manufacture, along with lesser quantities of buttermilk. Most of the skim milk was dried, and in some situations fed to animals. However, a further important trend over the last ten years has been the move to a more health-conscious diet, and in this sense skim milk is more widely used as the starting material for yoghurts, low-fat cheeses and other desserts. Ultrafiltered milk also forms the starting material for some of these types of product (de Boer and Koenraads, 1991). Cheese is a very important product derived from milk. In cheese production, most of the fat and the casein fractions are incorporated into the curd. However the by-product of cheese manufacture is whey, which incorporates the whey proteins (about 20% of milk protein). The compositions of whey and skim milk are given in Table 4.6. An IDF publication (International Dairy Federation, 1979) gives a summary of the rejection values obtained during the ultrafiltration of sweet whey, acid whey, skim milk and whole milk, using a series of industrial membranes. These results are summarised in Table 4.1. It should be noted that protein rejections are based on Kjeldahl nitrogen x 6.38. Where this is the case, it is also measuring non-protein nitrogen (NPN). Rejection values obtained may not be a true reflection of the behaviour of the protein. Such 122 M. J. Lewis rejection values could be lower than expected for materials containing substantial amounts of NPN. One example is chhana whey, which is produced from a heat- coagulated cheese and contains substantial NPN (Jindal and Grandison, 1992). However, low rejection values could also be indicative of excessive protein leakage, and may well warrant investigation if not expected. Rejection values based on true protein can be determined by Kjeldahl or by using electrophoresis or high performance liquid chromatography (HPLC) for analysis and will give a clearer picture of the behaviour of the proteins. Rejection values for ash are interesting and are higher than would be expected from their molecular weight. This would suggest that binding of minerals to the protein is occurring. During ultrafiltration of whey and buttermilk, Hiddink et al. (1978) observed that at pH 6.6, anions such as C1- and NO, were preferentially removed. However, at pH 3.2, cations such as Na', K', and Ca2+ were preferentially removed. This is an example of the Donnan effect. Maximum ash removal was obtained by ultrafiltration at pH 6.6, followed by diafiltration at pH 3 to 3.5. Bastian et 111. ( 199 I) compared the rejection values during ultrafiltration and diafiltra- tion of whole milk. They found that the rejection of lactose, riboflavin, calcium, sodium and phosphorus was higher during diafiltration than ultrafiltration. Diafiltration of acidi- fied milk gave rise to lower rejections of calcium, phosphorus and sodium. Premaratne and Cousin (1991) have performed a detailed study on the rejection of vitamins and minerals during ultrafiltration of skimmed milk. During a five-fold concentration the following minerals were concentrated by the following factors: Zn (4.9), Fe (4.9), Cu (4.7), Ca (4.3), Mg (4.0) and Mn (3.0), indicating high rejection values. On the other hand, most of the B vitamins examined were almost freely permeating. The use of nanofiltration or partial demineralisation has been discussed by Kelly et al. (1991) and its effects on lactose crystallisation, which was improved by about 8% at a concentration factor of 3, by Guu and Zall (1992). Both cheese whey and skim milk contain substantial protein and other nutrients. A great deal of attention has been paid to ultrafiltration of these products to increase their functionality and hence profitability. Cheese whey contains only about 10-12% protein on a dry weight basis. However the proteins are soluble and have excellent functional properties. The main thrust has been toward using ultrafiltration to increase the protein content, in order to produce concentrates, which could then be dried to produce high protein powders (concentrates and isolates) with useful functional properties. Some typical concentration factors (f) used are as follows: f'= 5 j'= 20 j'= 20; plus diafiltration protein content (dwb) about 35% (similar to skimmed milk) protein content about 65% protein content about 80% The product starts to become very viscous at a concentration factor of about 20. There- fore if a protein concentrate with a higher total solids is required, diafiltration is required. 1 .t 80- 124 M. J. Lewis Skim milk has been investigated also. However because of its original higher protein content, concentration factors of about 7 are the maximum achievable. Protein concentrates based on skim milk have not received the same amount of commercial interest as those based on whey proteins. However, it has been suggested that the concentrates can be further modified to produce an interesting range of products with good whipping and foaming characteristics. Rajagopalan and Cheryan (1991) reported that it was not possible to produce a pure protein isolate by ultrafiltration and diafiltration of skim milk, due to a high mineral rejection. An isolate containing about 90% protein and 8% ash was obtained. Yoghurt and other fermented products have been made from skim milk and whole milk concentrated by ultrafiltration (Renner and El-Salam, 199 1). Whey protein concen- trates have also been incorporated (de Boer and Koenraads, 1991). Production of labneh, which is a strained or concentrated yoghurt at about 21% total solids, has been described by Tamime et al. (1991), by preconcentrating milk to 21% TS. Inorganic membranes have also been used for skim milk, and Daufin et al. (1992) have investigated the cleanability of these membranes using different detergents and sequestering agents. As well as exploiting the functional properties of whey proteins, full cream milk has been concentrated by UF, prior to cheesemaking. The UF concentrate has been incorporated directly into the cheese vats. Some advantages of this process include: increased yield, particularly of whey protein, lower rennet and starter utilisation, smaller vats, or even complete elimination of vats, little or no whey drainage and better control of cheese weights. Lawrence (International Dairy Federation, 1989), suggests that concentration below a factor of 2 gives protein standardisation, reduced rennet and vat space, but no increased yield. At concentration factors greater than 2, an increased yield is found. Some problems result from considerable differences in the way the cheese matures and hence its final texture and flavour. The types of cheese that can be made in this way include: Camembert type cheese, mozzarella, feta and many soft cheeses. Those which are difficult include the hard cheeses such as Cheddar and also cottage cheese; the problems are mainly concerned with poor texture. Some debate about compositional standards for cheeses produced by such technology exists. More discussion is given by Kosikowski (1986). Further reviews on the technological problems arising during the conversion of retentate into cheeses are discussed by Lelievre and Lawrence (1988). Spangler et al. (1990a) have investigated how the quality of Gouda cheese was affected by concentration factor (3.6-5), rennet concentration and coagulation temperature. The best cheeses were made from the five-fold concentration and a rennet concentration of 0.14%. Higher rennet concentrations lead to the development of very bitter flavoured cheese. Spangler et al. (1990b), for example, found that Gouda production from ultrafiltered milk could be improved by attention to detail; such as preacidification of milk prior to ultrafiltration; the amount of rennet was also very important. Sachdeva and Reuter (1991) produced acceptable chhana by ultrafiltration; it had a lower moisture content, giving a yield improvement of 31.4% (product basis) and 16.4% on a dry matter basis. Quarg is also produced from ultrafiltered milk. Ultrafiltration I25 Whole milk has been concentrated five times by ultrafiltration and the retentate heated at 120°C for 5 min before being recombined with the permeate prior to pasteurisation (Kosikowski and Mistry, 1990). This procedure is claimed to produce an extended shelf- life product with a superior flavour to a conventionally pasteurised product. Sweetsuir and Muir (1985) have investigated the production of sterile concentrates, produced by the ultrafiltration of skim milk, with and without the addition of fat. Such concentrates were able to withstand sterilisation at 120°C for 7 min. The organoleptic qualities were improved by the presence of fat and the heat stability was improved by procedures which reduced the levels of salts in the concentrates. Protein hydrolysates produced from milk proteins have been discussed by Donnelly (1991). Ultrafiltration is an extremely valuable method of concentrating and recovering many of the minor components, particularly enzymes from raw milk, many of which would be inactivated by pasteurisation. Such enzymes are discussed in more detail by Kosikowski (1988). On-farm ultrafiltration processing of milk has been suggested as a way of reducing transportation costs, but is only viable for long distances, i.e. greater than 100 km, and is not likely in the UK; the permeate can be fed to pigs. Further reviews on membrane processing of milk are provided by Glover (1985, 1986), Renner and El-Salam (1991), El-Gazzar and Marth ( I99 1 ) and lnternational Dairy Federation ( 199 I ). 4.5.2 Oilseed and vegetable proteins Crops such as soyabean, cottonseed, sunflower and maize are gsown primarily for oil. After its extraction a solid residue remains which is often rich in protein. However, certain toxic components may also be associated with the residues. There have been many laboratory investigations into the use of ultrafiltration for extracting proteins from these residues, or removing the toxic residues. Lewis (1982) and Cheryan (1986) have reviewed the more important of these. For soya, the separation of low molecular weight peptides from soy hydrolysates, with the aim of improving quality; the dissociation of phytate from protein, followed by its removal by ultrafiltration; the removal of oligosaccarides; the removal of trypsin inhibitor and comparisons of performance of different membrane configurations. For cottonseed, the use of different extraction condi- tions has been evaluated, as have the functional properties of the isolates produced by ultrafiltration. Investigations were performed with sunflower and alfalfa to remove the phenolic compounds responsible for the colour and bitter flavour. Many have been very successful in their objectives and good quality protein concentrates and isolates have been produced, particularly with soyabean. For example, Nichols and Cheryan (1981) showed an 86% recovery of soya protein by a process involving multiple extraction, centrifugation, ultrafiltration and diafiltration. Diafiltration was necessary to produce an isolate, but there was still a relatively high amount of minerals, due to a higher than expected mineral rejection. mainly because of the economics of the process, dictated by the relatively low value of the products and the fact that acceptable food products can be obtained by simple However, in comparison to dairy processing, few have come to comme 126 M. J. Lewis technology, such as isoelectric precipitation. A further problem arises from the fact that the starting residue is in the solid form, thereby imparting an additional extraction proce- dure. Extraction conditions may need to be optimised, with respect to time, temperature, pH and antinutritional factors. Most vegetable proteins have their minimum solubility somewhere between pH 4 and 7. They are much more soluble at high and low pH values, so the use of second-generation membranes has helped enormously with processing such extracts and cleaning of such membranes. However, ultrafiltration at high pH and high temperatures may cause protein hydrolysis and excessive loss of nitrogen in the permeate. This has been observed by Lawhon et al. (1978). It is usually advisable to prefilter the extracts down to below 50 pm to avoid excessive flux decline due to fouling. A further problem arises from the complexity of oilseed and vegetable proteins, com- pared to milk products, evidenced by electrophoretic measurement. It is likely that many of these proteins are near their solubility limits after extraction and further concentration will cause them to come out of solution and promote further fouling. Fouling and clean- ing of membranes was found to be a serious problem during ultrafiltration of rapeseed meal extracts (Lewis and Finnigan, 1989). Again it was possible by extensive diafiltration to produce an extract with a high protein content and virtually no glucosinolate (dry weight basis). The rejection of glucosinolate was higher than expected, suggesting some form of binding with the proteins. Work has also been performed on potatoes and potato-processing wastes as well as bean and pea extracts. Waste streams from the vegetable industry have received attention, in order to recover starch and protein, as well as reducing biological oxygen demand (BOD). Some examples are given by Walters and Elliott (1983) and Koseoglu et al. (1991). Kosikowski (1986) has reviewed some of the work on clarification of alcoholic beverages and sugar juices from cane, beet and fruit rind. Fruit juices have been clarified successfully by membrane techniques; these are covered in more detail in Chapter 5. Vegetable juice processing is also important and is covered in more detail in the section on reverse osmosis (Chapter 3). Applications in cereal processing have been reviewed by Koseoglu (1991); these include the recovery of solids from thin stillage, protein from wet-milling processes and foam proteins from fermentation broths. A further important aspect is the use of enzyme reactors for the breakdown of starch to sugars. This is discussed in more detail later in this chapter tinder membrane reactors (Section 4.5.4). Oil-processing applications are very much in their infancy. Potential applications in vegetable oil production and refining are reviewed by Koseoglu (1991). One application is in the solvent-oil extraction process to recover the solvent from the oil, thereby reducing the amount of distillation required. This is possible using reverse osmosis and ultrafiltration membranes but the main problem is compatibility of the membrane with the hydrocarbon solvent, which can lead to collapse of the structure. For polysulphone membranes, this problem can be partially overcome by presoaking the membrane with solvents of decreasing polarity; 50/50 solution of isopropanol/water, followed by 50/50 solution of isopropanol/pentane, followed finally by pentane. Research and development has led to a new generation of membranes, which are more resistant to hydrocarbons, which not only remove solvent but also refine the oil by removing lecithin, free fatty acids, and the majority of the Ultrafiltration 127 coloured pigments. The membrane processing is claimed to eliminate degumming, alkali refining, water washing and drying. 4.5.3 Animal products Slaughterhouse wastes contain substantial amounts of protein. Two important streams that could be concentrated by ultrafiltration are blood and waste water. Blood contains about 17% protein. It can be easily separated by centrifugation into plasma (70%) and the heavier erythrocytes (red blood corpuscles or cells - 30%). Plasma contains about 7% protein, whereas the blood corpuscles contain between 28 and 38% protein. The proteins in plasma possess useful functional properties, particularly gelation, emulsification and foaming. They have been incorporated into meat products and have shown potential for bakery products, as replacer for egg white. Whole blood, plasma and erythrocytes have all been subjected to ultrafiltration processes (Ockerman and Hansen, 1988; Cheryan, 1986). The process is concentration polarisation controlled and flux rates are low. High flow rate and low pressure regimes are best. Gel concentrations were approximately 45% for plasma protein and 35% for red blood cells. The fouling characteristics of different blood fractions have been investigated by Wong et al. (1984). Whole blood was found to be the worst foulant, when compared to lysed blood and blood plasma. One significant problem with the erythrocyte fraction is the dark colour, due to haemoglobin, which affects its acceptability in most foods. Methods to reduce this include oxidation using hydrogen peroxide or separation of the haem group from the haemoglobin. However, this latter procedure reduces the stability of the protein. A third procedure involves digestion of the haemoglobin with proteolytic enzymes and the use of membranes to recover the breakdown products. Blood and other parts of the animal are used as raw materials for a wide range of active components. Waste water, like many other factory wastes, has a high BOD, as well as containing up to 10% of the incoming protein (Cheryan, 1986), and processes for removal of BOD and recovery of fat and protein are described by him. Another important material is gelatin, which can be concentrated from very dilute solutions by ultrafiltration. As well as concentration, ash is removed which improves its gelling characteristics. This is one example where there have been some high negative rejections recorded for calcium, when ultrafiltered at low pH. Eggs have also been processed by UF. Egg white contains 11-13% total solids (about 10% protein, 0.5% salts and 0.5% glucose). Large amounts of egg white are used in the baking industry. The glucose can cause problems during storage and causes excessive browning during baking. Whole egg contains about 25% solids and about 11% fat, whereas egg yolk contains about 50% solids. It is unusual to evaporate eggs prior to drying, because of the damage caused. Both RO and UF have been investigated as a means of concentrating egg prior to drying; UF also results in the partial removal of glucose; further removal can be achieved by diafiltration. Flux values during UF are much lower than for many other food materials, most probably due to the very high initial protein concentration; rates are also highly velocity dependent and temperature dependent (Cheryan, 1986). It is possible to remove any remaining glucose by fermentation, prior to 128 M. J. Lewis spray drying. Egg white can also be concentrated to about 20% total solids by RO; again, the remainder of the glucose can be removed by fermentation before being spray dried. There are various processes to remove proteins from eggs (lysosyme by ion exchange) after they have initially been concentrated by RO and UF. There is a great deal of interest in removing cholesterol from yolk, with strong marketing advantages for the resulting product, However, despite its relatively low molecular weight, this has not been achieved by UF. Extracts from cod offal have been subjected to ultrafiltration processes and three protein fractions were produced by a combination of ultrafiltration and pH precipitation. The extracts were hydrolysed but it was not found possible to further fractionate the hydrolysates by ultrafiltration (Vega and Brennan, 1987). 4.5.4 Biotechnology applications There is a range of applications within the field of biotechnology, concerned with the transformation of components by biocatalysts. These include microorganisms, enzymes and plant or animal cells. The range of substrates and products is vast and the desired product may be either the cells themselves or their metabolic by-products, which may be intracellular or extracellular. UF membranes have a role, both in the primary production process, for example as a reactor for fermentation processes or enzymatic reactions, or in downstream processing for recovering the products. Membrane-based bioreactors Membrane-based bioreactors appear to be a very promising application for the production of ethanol, lactic acid, acetone and butanol, starch hydrolysates and protein hydrolysates; further refinements include the introduction of electrodialysis and pervaporation, into selective processes. The two main types of membrane reactor are the enzyme reactor and the fermentation reactor. Enzyme reactors Many industrial processes involve the use of enzymes to break down molecules, for example the hydrolysis of starch or proteins, or of simpler molecules such as sucrose or lactose. Most such reactions involving enzymes are of a batch nature and the enzyme needs to be inactivated or is lost at the end of the process because it is not easy to separate it from the product. The processing costs of batch processes are also high. There is considerable interest in continuous processes, where the enzyme is either immobilised or retained within the reaction vessel and reutilised, whereas the products are removed and reutilised. Ultrathin membranes have been investigated for both these types of application, in the form of enzymatic reactors. Immobilisation of enzymes provides an alternative method to overcome some of the disadvantages of using free enzymes. One system which has been much investigated is the immobilisation of the enzyme by adsorption and entrapment onto the outside of tubular membranes. Hollow fibre systems have been found useful for this purpose, with the enzyme being immobilised on the shell side of the reactor (see Fig. 4.11(a)). This is useful in situations where the product and reactant are normally much smaller than the enzyme and both diffuse into the membrane from the feed. One example is the hydrolysis of lactose. These types of reactor are Ultrafiltration 129 Media Air - Permeate Reactor (a) (b) (c) Fig 4 I I Some ininiobilisation systems (a) immobiliadtion within the shell apace of a hollow tibre system, (b) continuous reactor linked to ultrafiltration module, for either enzymatic or fermentation reactions, (c) stirred cell or dead end reactor for enzymatic or fermentation reactions usually operated in plug flow mode. A variation on this theme is to immobilise the enzyme in the tubes and to pass the reactant through the shell; this is less common. The enzyme may also be immobilised within the membrane by incorporating it into the casting solution The second variant is where the enzyme is incorporated into the reaction vessel and the membrane not only physically retains the enzyme, but also allows a continuous removal of the product. This is not immobilisation in the strict sense of the word, but it does provide a mechanism for long-term usage. This arrangement is useful for hydrolysis of macromolecules, where the required end-product is often much smaller in molecular weight than the reactant or the enzyme. Dead-end or flow-through systems have been used (see Fig. 4.11(b, c)). The dead-end system uses the membrane unit both as the reaction unit and for separation and has been more widely investigated Flow-through systems use a cross-flow filtration unit connected in series with the reaction vessel. This arrangement has two major advantages over the dead-end system, namely better induced turbulence to ieduce concentration polarisation, which leads to poor activities in dedd-end reactors aftei only ii few hours’ operation and a separate location for the reaction and separation ve,sels, thereby permitting each to be operated at its optimum set of working conditions, rather than at a compromise set of conditions. Both systems operate essentially as a continuous stirred tank reactor (CSTR) and feed is supplied to replace the permeate removed. CSTRs are characterised by a wide distribu- tion of residence times and are usually considered to be well mixed, i e. the concentration of the feed to the membrane is the same as that in the reactor vessel. Thus at high conversions, the substrate concentration would be low; they are particularly useful for reactions which may be inhibited by substrate concentration; at first sight they may also appear useful because it helps reduce end-product inhibition. However, this is not necessarily the case because the end-product concentration in the permeate will be the same as that in the feed A further feature for first- and second-order reactions is that a large volume is required for a CSTR compared to a plug flow reactor The major advantages offered by this type of reactor are a high throughput and productivity, gov- erned by a high enzyme concentration attainable and a constant and long-term use of the enzyme 130 M. J. Lewis Cheryan (1986) has reviewed some of the potential applications for bioreactors for many types of protein and carbohydrate sources. Some other important aspects to be considered are the operating temperature and shear damage. Sometimes the optimum temperature for high activity may be one that also inactivates the enzyme quickly. Retained activity over a long period is an important function for membrane reactors. Shear damage to enzymes was once thought to be important, although there is little experimental evidence to suggest that it is a major cause of reduced activity over the short term. Denis et al. (1990) examined the effects of pumping on the activity of pectate-lyase and found no loss of activity after 7 h pumping, but 36% loss of activity after 25 000 passes over a period of 6 days. Loss of activity due to adsorption could be reduced by selecting a membrane with the same charge as the enzyme, at the prevailing shear. Vegetable proteins are more difficult to hydrolyse than animal proteins. The higher the levels of non-hydrolysable protein, the shorter will be the operational period for the reactor, due to the build-up of protein within the system. Cheryan and Deeslie (1983) reported run lengths of up to 90 h, during the hydrolysis of soya protein at 50"C, with yields between 85 and 94% and productivities (wt product/wt enzyme used) about 7 times higher than the equivalent batch process. There is scope for optimising the performance of such a reactor, in terms of temperature, enzyme concentration, substrate concentration, reactor volume and flow rate (also equivalent to permeate rate). Protein hydrolysates have many uses, such as flavour enhancers, improvement of functional properties, and in special dietary or medical applications, some of which have been recently reviewed by Donnelly (1991). Starch hydrolysis for the production of glucose syrups has also received some attention. There are two stages to the process, namely liquefaction and saccharification. Ideally, it would be desirable to perform both stages simultaneously. Liquefaction is difficult to achieve with enzymes at appreciable starch concentrations, because of the high viscosities involved (de Silva, 1991), although maltose has been produced by starch in a direct process (Nakajima et al., 1990). Saccharification, which makes use of glucoamylase enzymes, is more suited to an enzyme membrane process as the batch process can take up to 48 h and the substrate is less susceptible to concentration polarisation and fouling, due to its lower viscosity. Darnoko et al. (1989) examined the saccharification of cassava starch by glucoamylase in a dead-end stirred cell reactor. Essentially pure glucose was found in the permeate solids. Starch conversions were 64% at 25°C and 97% at 55°C. Productivity was 10-1 1 times better than a batch reactor over a 24 h period. de Silva (1991) has studied the saccharification of starch in a CSTR-recycle membrane reactor, using maltodextrins (DE 17) as the feed material, at concentrations between 10 and 30% total solids. A tubular membrane was used with a molecular weight cut-off of 4000. The performance of the reactor was evaluated for different feed concen- trations ( 10-30% TS), different enzyme concentrations and temperatures between 45 and 56°C. At 56°C and a feed concentration of 10% TS, a product containing 95% glucose was obtained, measured after 6 h operation. A mathematical model was developed to describe the process. This included the following terms: reaction kinetic parameters for pH .of the .l-paccjon, Kowc\;er, .&sue cwl.p&re and ari.r,.al cell.s are LTL3& m~re snyi.G.ve .to Ultrafiltration 131 the enzyme, including product inhibition, and temperature effects, on both reaction rate and enzyme inactivation; membrane permeability, which was described by the film theory model; and the rejection characteristics of the membrane, which were determined experi- mentally. However, there were slight losses in enzyme activity during processing. Other enzymic reactions which have been investigated include the hydrolysis of cellu- lose, soy protein, fish protein and casein (Cheryan, 1986). A semi-continuous process for lignocellulose hydrolysis has been described by Ishihara et al. (1991). In these long-term continuous processes it is important to minimise enzyme losses and in some processes, to retain, recover or regenerate important enzyme cofactors. Cofactors are required by many enzymes for activity and are classified as prosthetic groups, coenzymes and metal ions. They include compounds such as NAD, FAD, ATP, Co-A. Prosthetic groups are distinguished from coenzymes by their tighter binding to the enzyme, although there is considerable overlap in binding affinities of these two groups (Parkin, 1993). The cofactors may be immobilised or entrapped with the enzyme or retained within the system by selecting a membrane with a low molecular weight cut-off, which will totally reject them. Another approach is to use negatively charged membranes, which also give rise to higher rejections due to electrostatic interactions. Those which have mainly been investigated are the nicotinamide (NAD and NADP) cofactors. Another approach is to regenerate cofactors by coupled reactions. Examples investi- gated include mannitol production from fructose, by mannitol dehydrogenase. Here the NADH produced is regenerated by a coupled reaction involving a glucose dehydrogenase reacting on glucose to produce gluconic acid. The coenzyme turnover number was re- ported to be 150 000 (Kulbe et al., 1990). Membrane fermenters Fermentation processes can also be batch or continuous. Even with continuous processes, the conversion rate may be limited by end-product inhibition and the dilution rate must be less than the specific growth rate of cells, to avoid cell washout. Therefore, another application of membrane reactors is as part of a fermentation vessel, with containment of microorganisms and continuous removal of products. Again the microorganisms can be entrapped or immobilised, typically in the hollow fibre system, or simply allowed to circulate freely with the recycle broth. The freely circulating systems seem to perform more effectively than the immobilised systems. This system operates like a continuous system, with the additional feature that the cells are returned to the fermenter and retained within the system. It is possible to achieve high cell densities and high dilution rates, without the worry of washing out the cells. However, they have not been as thoroughly investigated as enzyme reactors. One of the most popular applications is the production of alcohol. Cheryan (1986) compares the performance of batch fermentation processes, with continuous culture, immobilised cells and membrane recycle. The membrane recycle gives the biggest productivity in terms of mass of ethanol per unit volume and time. He has also described the use of a membrane reactor to produce alcohol from concentrated whey permeate, using a very high cell density which is retained by the membrane. Such a system has shown a high productivity and has been operated continuously for up to 10 days. 132 M. J. Lewis Long-term stability needs to be established. Teixeira et al. (1990) used lactose as the carbon source and found that ethanol productivity was 12.5 times higher than continuous fermentation and was directly proportional to the dilution rate. Biomass concentration showed a linear relationship with dilution rate and the largest concentration was eight times that obtained in a conventional continuous fermenter. Endo et al. (1990) describe a tubular bioreactor for high-density cultivation of microorganisms, used in the primary part of the circuit. They found improved cell density by over 10 times compared with a conventional fermenter and improved productivity compared to operating at constant dilution rate. Crespo et al. (1990) studied the effects of recirculation rate for a propionic acid bacterium fermentation process, linked to a tubular UF module, on permeate flux, fermenter homogeneity, energy consumption, temperature rise and cell damage. High circulation rates improved flux and the degree of homogeneity, but increased cell damage. Energy input was also estimated to determine the degree of cooling required. Propionic bacteria are useful for the production of propionic acid, cyanocobalamin, flavours and dairy starter cultures. Minier et al. (1990b) combined the membrane process with distillation, and found that the combination was useful to further concentrate the alcohols and to remove the low molecular weight inhibitors. A mineral ultrafiltration membrane has been used to separate lytic enzymes produced extracellularly during an acetonobutylic fermentation process. The membrane retained the cells but allowed the lytic enzymes to permeate. The permeate was heat treated to inactivate the enzymes, before being returned to the fermenter (Minier et al., 1990a). However, although lytic enzyme activity was decreased by over 6 times, there was no marked increase in the yield of butanol compared to the control. A novel method of immobilisation involves sandwiching the cells between a reverse osmosis and ultrafiltration membrane. The UF membrane allows free passage of the nutrients to the cells. The RO membrane helps immobilise the cells and permits separation of the alcohol from the sugars and salts. This system has been investigated experimentally and modelled by Jeong et al. (1991). There also exists the possibility of cultivation of animal and plant cells in bioreactors. The performance of a hollow fibre immobilised system has been investigated for the production of antibodies (Piret and Cooney, 1990). Procedures which reduced cell and protein distribution on the shell side of the hollow fibre membrane, led to an improve- ment in productivity. Ultrafiltration membranes provide a number of advantages, including the retention of growth factors and the selective concentration of high molecular weight protein products. Recovery of components and downstream processing Ultrafiltration is now widely used for the recovery of cells, enzymes and other metabolic products. Ultrafiltration and microfiltration can be used for concentrating and harvesting cells but will be in direct competition with continuous centrifugation and to a lesser extent, conventional filtration. The cells may also be shear sensitive, so this factor needs some consideration. Factors such as membrane geometry, pore size, chemical nature of the membrane and ultrastructure should be investigated. Fouling is not necessarily related only to pore size, but to the distribution of pore size, the relationship between pore size and particle size, and whether fouling occurs on the surface or within the pores. Cheryan Ultrafiltration 133 (1986) concludes that the overall economics favour membrane processes for small-scale operations, for all particle sizes. However, for larger-scale operations, the economics may favour centrifuges with larger particles, e.g. yeast cells, but the membrane systems for smaller particles. Ultrafiltration is widely used commercially for concentrating enzyme solutions. There are also many diafiltration operations, for washing out low molecular weight contami- nants. For example, Olsen et al. (1990) describe the recovery of four enzymes, alkaline phosphatase, hyaluronoglucuronidase, chitinase and beta-N-acetylglucosaminidase, from shrimp waste. Flocculation was induced by ferric chloride at pH 7. The supernatant was then ultrafiltered to recover 99% of the enzyme activity. However, fractionation of the enzymes was not investigated. Pacheco-Oliver et al. (1990), used diafiltration to remove coloured impurities from a lipase enzyme, produced by fermentation. There has been much less success for fractionating enzymes or other components, based solely on size difference, unless there is at least a tenfold difference in their molecular weights. Again the selection of membranes is very important, to minimise excessive enzyme loss in the permeate. However, there has been some success at partial fractionation and purification, by using compounds which may either bind with one of the fractions and alter its molecular weight and hence its rejection characteristic, or by contacting it with a specific compo- nent with which it has a strong affinity. This other component could be introduced in the feed or immobilised within hollow fibre membrane systems. Thus, in conjunction with affinity or ion exchange gels or resins, a specific component in the crude extract can be bound and separated, and subsequently eluted. Examples are given for the purification of horse serum cholinesterase and cattle liver carboxylesterase by these types of method, using an affinity gel and DEAE sephadex, respectively (Molinari et al., 1990). Trypsin has been purified by affinity chromatography and ultrafiltration, using a water-soluble ligand-bound polymer. This was incubated with the crude extract, forming a soluble polymer-trypsin complex. Unbound enzymes were removed by ultrafiltration, whereas the complex was retained. Trypsin was then eluted from the polymer and the eluant was subject to ultrafiltration to further purify the trypsin (US patent, 1990). A French patent (anon) (1991) describes the isolation and recovery of a wide range of enzymes from waste water treatment sludge. Ultrafiltration is used to concentrate the liquid phase, which results from the sludge treated by a number of methods. Another development is the attachment of bacteria with outer layers rich in glycoproteins to microfiltration membranes. These are then used as a support for protein immobilisation, or as an ultrafiltration membrane (Sara et al., 1990). It is also possible to separate components such as dissolved organic materials or metal ions by micellar enhanced ultrafiltration (MEUF). The compounds are solubilised within micelles by the addition of surfactants and subsequently removed by ultrafiltration. The processes and factors affecting the flux and rejection of surfactants during ultrafiltration and reverse osmosis are reviewed by Akay and Wakeman (1993). 4.5.5 Medical applications: serum fractionation One important medical application is for treatment of patients suffering kidney failure, which leads to an accumulation of toxic components, such as urea and creatine within the 134 M. J. Lewis blood. There are several treatment choices available: peritoneal dialysis, haemodialysis, haemofiltration and haemodiafiltration (de Burgh, 1992). Peritoneal dialysis makes use of the natural membrane in the peritoneum. A glucose solution is infused into the peritoneal cavity, and solutes and water from the blood diffuse through this membrane into the sugar solution. The sugar solution is infused and removed at hourly intervals. No direct access to the blood supply is required. The other three applications use a manufactured membrane to achieve the removal, with blood being supplied to the membrane. Haemodialysis is usually a short-term, intermittent process, of 2-4 h duration, which requires the availability of dialysis trained nursing staff, whereas haemofiltration and haemodiafiltration can be used continuously for up to several days. Haemodialysis in- volves the removal of plasma water, containing toxins, through the membrane filter, which is usually either a flat plate or hollow fibre configuration. This filtrate is replaced with a sterile infusion fluid to maintain the patient’s fluid status. Where sufficient blood pressure is present, the blood flow can be driven by the patient’s own arterial pressure, using either an arterio-venous shunt or an arterial groin line. Where blood pressure is insufficient, a blood pump may be incorporated. Whichever technique is used, the utmost care should be taken to ensure that blood flow is maintained and the filter does not become blocked. Haemodiafiltration uses the same blood circuit, but incorporates a dialysate solution, which is passed through the filter on the permeate side of the mem- brane, thereby facilitating toxin removal by diffusion. Control of fluid removal may be achieved by adjusting the output of dialysate from the filter in relationship to its input flow. The operating conditions and performance of these two systems are reviewed by Kox and Davies (1992). Further discussion on the use of membranes in artificial organs is given by Leonard (1987). REFERENCES Aimar, P. (1 987) Ultrafiltration of pseudoplastic fluids. In Separarions for Biotechnology, Verrall, M. S. and Hudson, M. J. (eds.), Ellis Horwood, Chichester, pp. 360-372. Akay, G. and Wakeman, R. J. (1993) Ultrafiltration and microfiltration of surfactant dispersions - an evaluation of published research, Trans. Itist. Chern. Eng. 71, Part A, 41 1-420. Ali Asbi, B. and Cheryan, M. (1992) Optimizing process time for ultrafiltration and diafiltration, Desalination, 86, 49-62. Amicon (1992), Laboratory Products Catalogue, Stonehouse, Glos. Bastian, E. D., Collinge, S. K. and Emstrom, C. A. (1991) Ultrafiltration; partitioning of milk constituents into permeate and retentate, J. Dairy Sci., 74, 2423-2434. Bohner, H. F. and Bradley, R. L. Jr. (1992) Effective cleaning and sanitizing of polysulphone ultrafiltration membrane systems. J. Dairy Sci., 75, 7 18-724. Cheryan, M. (1986) Ultrafiltration Handbook, Technomic Publishing Co., Lancaster, USA. Cheryan, M. (1989) Membrane separations: mechanisms and models. In Food Properties and Computer Aided Engineering of Food Processing Systems, Singh, R. P. and Medina, A. G. (eds.), Kluwer, Amsterdam. Ultrafiltration 135 Cheryan, M. and Deeslie, W. D. (1983) Soy protein analysis in membrane reactors, J. Amer. Oil Chem Soc., 60, 11 12-1 115. Colman, D. A. and Mitchell, W. S. (1991) Enhanced mass transfer for membrane proc- esses, Trans. liist. Chem. Eng., 69, Part C, 91-96. Coolbear, T., Monk, C., Peek, K., Morgan, H. W. and Daniel, R. M. (1992) Laboratory- scale investigations into the use of extremely thermophilic proteinases for cleaning ultrafiltration membranes fouled during whey processing, J. Membrane Sci., 67, 93- 101. Crespo, J. P. S. G., Moura, M. J. and Carrondo, M. J. T. (1990) Some engineering parameters for propionic acid fermentation coupled with ultrafiltration, Appl. Biochem. Biotechnol., 24, 61 3-625. Darnoko, M., Cheryan, M. and Artz, W. E. (1989) Saccharification of cassava starch in ultrafiltration, Enz. Microbial Technol., 11, 154-159. Daufin, G., Labbe, J. P., Quemerais, A. and Michel, F. (1991) Fouling of an inorganic membrane during ultrafiltration of defatted whey protein concentrates, Netherlands Milk and Dairy J., 45, 259-272. Daufin, G., Merin, U., Kerherve, F. L., Labbe, J. P., Quemerais, A. and Bousser, C. (1992) Efficiency of cleaning agents for an inorganic membrane after milk ultrafiltra- tion, J. Dairy Res., 59, 29-38. de Boer, R. and Koenraads, J. P. J. M. (1991) Incorporation of liquid ultrafiltration - whey retentates in dairy desserts and yoghurts. In New Applications in Membrane Processes, IDF Special Issue, No. 9201. de Burgh, J. (1992) The nursing management of acute renal failure, Care ofthe Critically Denis, S., Terre, S., Bertheau, Y. and Boyaval, P. (1990) Factors affecting pectate-lyase activity during membrane filtration, Biotechnol. Tech., 4, 127-132. de Silva, M. (1991) Optimisation of a continuous membrane bioreactor for the saccharification of starch using temperature as a control variable: experimental study and simulation analysis. PhD thesis. University of Reading, UK. Donnelly, W. J. (1991) Applications of biotechnology and separation technology in dairy processing, J. Soc. Dairy Technol., 44, 67-72. El-Gazzar, F. E. and Marth, E. H. (1991) Ultrafiltration and reverse osmosis in dairy technology; a review, J. Food Protection, 54, 801-809. Endo, I., Nagamune, T., Taichikawa, S. and Inabe, H. (1990) A tubular bioreactor for high density cultivation of micro-organisms, Adv. Biochem. Eizg. Biotechnol., 42, 1- 11. Fane, A. G. and Fell, C. J. D. (1987) A review of fouling control of ultrafiltration, Desalination, 62, 117-136. French Patent (1991) Patent No. FR2650597. Gekas, V., Traghard, G., Capannelli, G. and Bottino, A. (1990) Correlation of direct porosimetric data and performance of ultrafiltration membranes, Process Biochern. Glover, F. (1985) Ultrafiltration and reverse osmosis for the dairy industry. Technical Glover, F. (1986) Modifications to the composition of milk. In Modern Dairy 111, 8, 15-17. ltzt., 25, 11 1-1 16. Bulletin No. 5, NIRD, Reading, England. 136 M. J. Lewis Technology, Vol. 1, Advances in milk processing, Robinson, R. K. (ed.), Elsevier Applied Science Publishers, London. Guu, Y. K. and Zall, R. R. (1992) Nanofiltration concentration on the efficiency of lactose crystallisation, J. Food Sci., 57, 735-739. Hayes, J. F., Dunkerley, J. A,, Muller, L. L. and Griffin, A. T. (1974) Studies on whey processing by ultrafiltration: improving permeation rates by preventing fouling, Aust. J. Dairy Res., 29, 132-140. Heng, M. H. and Glatz, C. E. (1991) Chemical pretreatments and fouling in acid cheese whey ultrafiltration, J. Dairy Sci., 74, 11-19. Hiddink, J., de Boer, R. and Romijn, D. J. (1978) Removal of inilk salts during ultrafiltra- tion of whey and buttermilk, Neth. Milk Duirji J., 32, 80-93. Howell, J. A,, Wu, D. X. and Aimar, P. (1990) Entrance length of the concentration boundary layer in ultrafiltration and its effect on scale-up. BIUTEC-90, 353-358. International Dairy Federation (1979) Bulletin No. 115. International Dairy Federation (1991) New Applications of membrane processes. Special Issue No. 9201. Ishihara, M., Uemura, S., Hayashi, N. and Shimizu, K. (1991) Semi-continuous enzy- matic hydrolysis of lignocellulose, Biotechnol. Bioeng., 37, 948-954. Jeong, Y. S., Vieth, W. R. and Matsuura, T. (1991) Transport and kinetics in sandwiched membrane bioreactors, Biotechnol. Prod., 7, 130-139. Jindal, A. R. and Grandison, A. S. (1992) Preparation and composition of chhana whey powders using membrane techniques, J. Sci. Food Agric., 58, 5 1 1-5 17. Kelly, P. M., Horton, B. S. and Burling, H. (1991) Partial demineralization of whey by nanofiltration. In New Applications of Membrane Processes, IDF Special Issue No. 9201. Kessler, H. G., Gernedel, C. and Nakanishi, K. (1982) The effect of low molecular weight milk constituents on the flux in ultrafiltration, Milchwissenschafi, 37, 584-587. Koseoglu, S. S. (1991) Membrane technology for edible oil refining, Oils urzd Fats Irzternational, 7, 16, 18, 20-21. Koseoglu, S. S., Rhee, K. C. and Lusas, E. W. (1991) Membrane separations and applica- tions in cereal processing, Cereal Food World, 36, 376-383. Kosikowski, F. V. (1986) Membrane separations in food processing. In Menibrune Sepa- rations i/7 Biotechnology, McGregor, W. C. (ed.), Marcel Dekker, New York. Kosikowski, F. V. (1988) Enzyme behaviour and utilisation in dairy technology, J. Dairji Sci., 71, 557-573. Kosikowski, F. V. and Mistry, V. V. (1990) Microfiltration, ultrafiltration, centrifugation separation and sterilisation processes for improving milk and cheese quality. J. Dairy Sci., 73, 141 1-1419. Kox, W. J. and Davies, S. P. (1992) Continuous dialysis in the critically ill, Care of Critically 111, 8. 8-1 1. Kulbe, K. D., Howaldt, M. W., Schmidt, K., Roethig, T. R. and Chmiel, H. (1990) Rejection and continuous regeneration of native coenzyme NAD(P)H in a charged ultrafiltration membrane reactor, Ann. N. Y. Acad. Sci., 613, 820-826. Kuo, K.-P. and Cheryan, M. (1983) Ultrafiltration of acid whey in a spiral wound unit; effect of operating parameters on membrane fouling. J. Food Sci., 48, 11 13-1 118. Ultrafiltration 137 Lawhon, J. T., Hensley, D. W., Mulsar, D. and Matill, K. K. (1978) Optimization of protein isolate production from soy flour, using industrial membrane systems, J. Food Lawrence, R. C. (1989) The use of ultrafiltration technology in cheese making, Znterna- tional Dairy Federation, Bulletin No. 240. Lee, D. N. and Merson, R. L. (1976) Chemical treatment of cottage cheese whey to reduce fouling of ultrafiltration membranes, J. Food Sei., 41, 778-786. Lelievre, J. and Lawrence, R. C. (1988) Manufacture of cheese from milk concentrated by ultrafiltration, J. Dairy Sei., 55, 465-478. Leonard, E. F. (1987) The applications of artificial organs to biotechnology. In Mem- brane Separutions in Biotechnology, McGregor, W. C. (ed.), Marcel Dekker, New York, pp. 319-353. Lewis, M. J. (1982) Ultrafiltration of proteins. In Developments in Food Proteins, Vol. 1, Hudson, B. J. F. (ed.), Applied Science Publishers, London, pp. 91-130. Lewis, M. J. and Finnigan, T. J. A. (1989) Removal of toxic components using ultrafiltra- tion. In Process Engineering in the Food Industry, Field, R. W. and Howell, J. A. (eds.), Elsevier Applied Science, London, pp. 291-306. McGregor, W. C. (1986) Selection and use of ultrafiltration membranes. In Membrane Separations in Biotechnology, McGregor, W. C. (ed.), Marcel Dekker, New York, pp. Minier, M., Soucaille, P., Blanc-Ferras, E., Izard, A. and Goma, G. (1990a) Inactivation of lytic enzymes by heat treatment in ultrafiltration-coupled acetonobutylic fermenta- tion, Appl. Microbiol. Biotechnol., 33,274-279. Minier, M., Grateloup, R., Blanc-Ferras, E. and Goma, G. (1990b) Extractive acetonobutylic fermentation by coupling ultrafiltration and distillation, Biotechnol, Bioeng., 35, 861-869. Molinari, R., Torres, J. L., Michaels, A. S., Kilpatrick, P. K. and Carbonelli, R. G. (1990) Biotechnol. Bioeng., 36, 572-580. Nakajima, M., Iwasaki, K., Nabetani, H. and Watanabe, A. (1990) Continuous hydrolysis of soluble starch by free beta-amylase and pullanase using an ultrafiltration membrane reactor, Agric. Biol. Chem., 54,2793-2799. Nichols, D. J. and Cheryan, M. (1981) Production of soy isolates by ultrafiltration: factors affecting yield and composition, J. Food Sei., 46, 367-372. Nisbett, T. J., Thorn, T. M. and Wood, P. W. (1981) Observations on the fouling of polysulphone membranes by acid whey, New Zealand J. Dairy Sei. and Technol., 16, Ockerman, H. W. and Hansen, C. L. (1988) Animal By-product Processing, Ellis Hor- wood, Chichester. Olsen, R. L., Johansen, A. and Myrens, B. (1990) Recovery of enzymes from shrimp waste, Process Biochem., 25, 67-68. Pacheco-Oliver, M., Veeraragavan, K. and Braendli, E. (1990) Separation of colour com- pounds from lipase in fermentation supernatant by diafiltration. Biotechnol. Tech., 4, Parkin, K. L. (1993) General characteristics of enzymes. In Enzymes in Food Processing, Sci., 43, 361-364, 369. 1-35. 113-120. 369-372. Nagodawithana, T. and Reed, G. (eds.), Academic Press, San Diego, pp. 7-37. 138 M. J. Lewis Patocka, J. and Jelen, P. (1987) Calcium chelation and other pretreatments for flux improvement in ultrafiltration of cottage cheese whey, J. Food Sci., 52, 1241-1244. Piret, J. M. and Cooney, C. L. (1990) Mammalian cell and protein distributions in ultra- filtration hollow fibre bioreactors, Biotechnol. Bioeng., 36, 902-910. Premaratne, R. J. and Cousin, M. A. (1991) Changes in the chemical composition during ultrafiltration of skim milk, J. Dairy Sci., 74, 788-795. Rajagopalan, N. and Cheryan, M. (1991) Total protein isolate from milk by ultrafiltra- tion; factors affecting product composition, J. Dairy Sci., 74, 2435-2439. Ramachandra Rao, H. G., Lewis, M. J. and Grandison, A. S. (1994) Effect of soluble calcium of milk on fouling of ultrafiltration membranes, 1. Sci. Food Agric., 65, 249- 256. Renner, E. and El-Salam, M. H. Abd. (1991) Application of Ultrafiltration in the Dairy Industry, Elsevier Applied Science, London. Sachdeva, S. and Reuter, H. (1991) Production of chhana by ultrafiltration, J. SOC. Dairy Technol., 44, 95-98. Sara, M., Kupchu, S. and Sleytre, U. B. (1990) Crystalline bacterial cell surface layers used as ultrafiltration membranes and immobilisation matrix, Genet. Eng. Biotechnol., Sirkar, K. K. and Prasad, R. (1987) Protein ultrafiltration - some neglected considera- tions. In Membrane Separations in Biotechnology, McGregor, W. C., (ed.), Marcel Dekker, pp. 37-59. Spangler, P. L., Jensen, L. A,, Amundsen, C. H., Olson, N. F. and Hill Jr., C. G. (1990a) Gouda cheese made from ultrafiltered milk; effects of concentration factor, rennet concentration and coagulation temperature, J. Dairy Sci., 73, 1420-1428. Spangler, P. L., Jensen, L. A,, Amundsen, C. H., Olson, N. F. and Hill Jr., C. G. (1990b) Ultrafiltered gouda cheese; effects of pre-acidification, diafiltration, rennet and starter concentration and time to cut. J. Dairy Sci., 74, 2809-2819. Suki, A,, Fane, A. G. and Fell, C. J. D. (1986) Modelling fouling mechanisms in protein ultrafiltration, J. Membrane Sci., 27, 181-193. Sweetsuir, A. W. M. and Muir, D. D. (1985) Effect of fat incorporation on properties of sterile concentrates prepared by ultrafiltration of skimmed milk, J. SOC. Dairy Technol., 38, 88-93. Tamime, A. Y., Davies, G., Chekade, A. S. and Mahdi, H. A. (1991) The effect of processing temperatures on the quality of labneh made by ultrafiltration, J. Soc. Dairy Technol., 44, 99-103. Teixeira, J. A,, Mota, M. and Goma, G. (1990) Continuous ethanol production by a flocculating strain of Kluyveromyces marxianus: bioreactor performance, Bioprocess Eng. (1990), 5, 123-127. Thomas, R. L., Cordle, C. T., Criswell, L. G., Westfall, F. H. and Barefoot, S. F. (1992) Selective enrichment of proteins using formed-in-place membranes, J. Food Sci., 57, 10, 10-13. 1002- 1005. US Patent (1990) Patent No. US4973554. Vega, R. E. and Brennan, J. G. (1987) Fractionation of fish proteins by ultrafiltration. In Separationsfor Biotechnology, Verall, M. S. and Hudson, M. J. (eds.), Ellis Horwood, Chichester, pp. 373-377. Ultrafiltration 139 Walstra, P. and Jenness, R. (1984) Dairy Chemistry and Physics, Wiley, New York. Walters, J. K. and Elliott, K. L. (1983) Ultrafiltration in the recovery of food wastes. In Upgrading Wastes for Feeds and Foods, Ledward, D. A., Taylor, A. J. and Lawrie, R. A. (eds.), Butterworths, London, pp. 51-72. Wong, W., Jelen, P. and Chang, R. (1984) Ultrafiltration of bovine blood. In Engineering and Food, Vol. 1, McKenna, B. (ed.), Elsevier Applied Science, London, pp. 551-558. Wu, D., Howell, J. A. and Turner, N. M. (1991) A new method for modelling the time- dependence of permeate flux in ultrafiltration, Trans. Inst. Chem. Eng., 69, Part C, 77- 81.