Commercial spray-dried WPI from cheese whey, Provon 190, was purchased from Glanbia Nutritionals Inc. (Richfield, ID, USA) and contained 90.1% (w/w) protein, with 3.6% moisture, 2.9% ash, and lactose and fat. The approximate protein composition as determined with a combination of electrophoresis and chromatography methods [42,44] was: 55% β-LG, 20% α-LA, 18% GMP and 7% minor whey proteins (immunoglobulins, bovine serum albumin, lactoferrin, and casein fragments).

WPI solutions with concentrations, C_WPI = 50, 100 or 150 g/L (5, 10 or 15 wt %) were prepared using de-ionized water produced with a Milli-Q Synthesis water purification system (Millipore, Billerica, MA, USA). Liquid carbon dioxide (CO₂) tanks with an eductor tube were purchased from GTS-Welco (Allentown, PA, USA).

A pilot-scale batch process to fractionate the proteins of WPI with sCO₂ was adapted from previous work by Tomasula et al. [33,41,45,46]. A 1-L stainless-steel, high pressure reactor from Autoclave Engineers (Erie, PA, USA), with bolted closure, floor stand, MAG 075 MagneDrive stirring assembly, and a maximum working pressure of 38 MPa was used to saturate 700-mL WPI solutions with supercritical CO₂ (sCO₂), as shown in Figure 1. Various accessories were included or added for control of the temperature, pressure, sampling, and safety of operation.

Liquid CO₂ was pumped into the reactor and pressurized to supercritical conditions with an air-driven liquid pump (Haskel, Huntington Beach, CA, USA). The solution and sCO₂ were mixed with a turbine impeller mounted on a MAG075 MagneDrive stirrer (Autoclave Engineers, Erie, PA, USA) set to 1000-rpm. A turbine impeller allowed fast (3 to 7 min) thermodynamic saturation of the liquid with CO₂ through the dispersion and re-circulation of CO₂ bubbles taken from the reactor’s headspace into the protein solution. The reactor was heated slowly to the desired temperature, T_R = 60 to 65 °C, with an electric heating mantel (Autoclave Engineers, Erie, PA) and the solution temperature was measured with a thermocouple dipped in a well inside the reactor and recorded on a computer with a programmable PID temperature-controller (Cole-Parmer, Vernon Hills, IL, USA). The reactor was pressurized with liquid CO₂ to 5.5 MPa before heating, then additional CO₂ was pumped into the warm reactor to reach the desired pressure, P = 8.3 or 31 MPa. Steady-state for T and P was monitored with the temperature-controller, pressure transducer and LabView software (National Instruments Corp., Austin, TX, USA) during the duration of the experiment (2 hours). After treatment, the protein mixture was extracted through a dip-tube, using the pressure inside the reactor as the driving force. During depressurization, vaporization of dissolved CO₂ cooled the sample and produced dense, white foam. After foam collapse, pH of the sample returned to ~6.0. The sCO₂-treated protein mixture was centrifuged at 10,000 g with a Sorvall RC-5B centrifuge with a SA-600 or SS-34 rotor, then the solid and liquid fractions were separated and lyophilized. The reactor base was sterilized with steam then thoroughly washed between runs. Each experiment was replicated with 2 to 4 repeats and data were analyzed using ANOVA on MS Excel 2003 software (Microsoft, Redmond, WA, USA).

In aqueous solutions, CO₂ produces carbonic acid when dissolved, lowering the pH as a function of C_WPI, P and T according to thermodynamic equilibrium [43] and buffering effects from the whey proteins. The pH of WPI solutions saturated with CO₂ as a function of C_WPI and P was extensively measured, then calibrated for calculation as a function of C_WPI and P (unpublished results). Briefly, the pH of WPI solutions ranging from 10 to 280 g/L was measured at 60 °C between 0.5 and 13.8 MPa with a high pressure probe resistant to 13.8 MPa (Innovative Sensors, Inc., Anaheim, CA, USA), then modeled as a function of P and C_WPI with Equations 1 and 2 in order to calculate pH values at higher pressures:

with P in psi, and

valid for C_WPI = 1 to 28 wt % WPI.

The aggregate fraction yield, Y, was determined by mass difference as the ratio of solid material in the dried precipitate to the initial amount of WPI powder.

Individual recovery of protein k in the aggregate fraction, rec_k, was calculated as the mass ratio of protein k recovered in the precipitate, divided by the amount of protein k in the initial WPI solution:

where m_agg and m_WPI are the total mass of precipitate (aggregate) and WPI powder, respectively; and x_k,agg and x_k,WPI are the mass fractions of protein k in the aggregate fraction and the initial WPI powder, respectively, as determined via compositional analyses.

The protein distribution in both fractions was determined using sodium-dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) on a Phast System (Pharmacia, Piscataway, NJ) with 20% acrylamide homogeneous gels and 8 lanes. Samples from all the solid and liquid fractions were prepared similarly to the method of Parris et al. [44], using 7 wt % protein into a 2.5% SDS solution. Gel treatment included staining of protein bands with Coomassie brilliant blue dye for 20 min, and de-staining for 4 hours. Location, size and intensity of the protein bands were characterized with ImageQuaNT^TM software (Molecular Dynamics, Inc., Sunnyvale, CA, USA). The composition of the protein fractions was obtained using the software’s automated ‘volume %’ calculation for each protein band (the color intensity integrated over the surface area of the band) and corrected with a calibration curve.

Because GMP was not visible on SDS-PAGE gels, several samples were also analyzed with high-pressure liquid chromatography (HPLC) to measure GMP contents and adjust compositions accordingly. HPLC was run on a 250 mm long C4 column with 3.5 μm particles and 3 × 10⁻⁸ pores, with acetonitrile solvent containing 0.1% TFA at a flow rate of 0.8 mL/min, and UV detection set at 214 nm wavelength [42]. Protein solutions were diluted to 1 wt % concentration.

The experimental results were used to develop a process flow diagram, flow-sheet and cost estimation models of a scaled-up, semi-continuous version of the sCO₂ whey protein fractionation process. WPI at a rate of 200 pounds per hour (90.7 kg/h) mixed with water to form a solution of 10% or 5% WPI concentration, was used as the incoming feed, and two enriched whey protein powders as the final products.

A flow-sheet of the various streams at different points of the process was developed in agreement with experimental results and simple mass balances with Microsoft Excel software.

The SuperproDesigner® (Intelligen, Inc., Scotch Plains, NJ, USA) computer process simulation and cost analysis program, together with mass-balance data from the flow-sheet, were employed to design a commercial-scale process flow diagram, characterize the various process flows and equipment, and aid in the development of the capital and operating costs.

Capital costs were estimated with the SuperproDesigner® simulation and costing module. Costs of all major process equipment (including high-pressure reactor, high-speed centrifuge, spray-dryer, drum dryer, compressor) were based on information received from equipment suppliers and from in-house estimating sources. The total plant installed cost was calculated from total equipment costs through the use of Lang factors [45]. A total plant cost of three times the estimated cost of facilities equipment included: supply and installation of the process equipment and all support materials such as piping, electrical, instrumentation; foundations and buildings to house the equipment; facilities design and construction management; start-up expenses.

The production time was fixed as 22 hours/day, 250 days/year. Process operating costs included raw materials (WPI: $10.70/kg in July 2011), water ($0.71/m³), steam ($12/ton), electricity ($0.10/kWh), maintenance time (2 hours/day), maintenance cost (6% of capital cost per year), operators (two per shift, $45/hour), and insurance (1% of capital cost per year). We assumed full recycling of CO₂.

To choose an industrial-size centrifuge, the influence of centrifuge speed on the general compaction of α-LA aggregated particles and the retention of β-LG-rich solution between and within the particles was studied as a function of initial solution concentration (C = 5, 10 and 15wt %), due to the assumption that protein concentration will affect the viscosity of the WPI solution and thus change the viscous drag in effect during particle sedimentation. Identical 40-mL samples of sCO₂-treated WPI (as described above) were centrifuged at 4 different speeds for 30 min. The force in the middle of the tube was chosen as the following: 2000 g, 5000 g, 10,000 g and 20,000 g (corresponding to 5000 rpm, 8000 rpm, 11,500 rpm and 16,000 rpm according to rotor specifications, respectively).

To design the commercial-size de-foaming tank needed at the reactor exit (after depressurization of the sCO₂-treated protein mixture), we studied some of the general foaming properties of the sCO₂-saturated WPI solution. WPI solutions of various concentrations (20 to 150 g/L) were saturated with CO₂ and equilibrated for 15 min at 60 °C and various values of P (4 to 31 MPa), to study the effect of solubilized-CO₂ and protein contents on the quality of the foam. Large foam samples (500 mL) were extracted through the dip-tube and general foam attributes (initial density, liquid drainage rate and total lifetime) were recorded.

The supercritical CO₂ process effectively enabled fractionation of the proteins of WPI into a solid, α-LA-enriched fraction and a soluble, β-LG-enriched fraction, via precipitation of α-LA under acidic conditions between pH 4.4–5 at 60 to 65 °C and under CO₂ pressures of 5.5 to 31 MPa for 2 hours. Both fractions contained no salt, acid or other contaminant after processing, and had a final pH of 6.0. Figure 2 shows the typical protein distribution of the precipitate fraction after centrifugation and drying, as measured with SDS-PAGE and analyzed with densitometry: peaks 1 through 6 correspond to lactoferrin, bovine serum albumin and immunoglobulins; peaks 7, 8 and 11 are most likely caseins and casein fragments; and peaks 9 and 10 correspond to β-LG and α-LA, respectively. The double peak for β-LG is due to the two different conformations of β-LG [13].

The initial WPI contained almost three times more β-LG than α-LA, and a small amount of minor whey proteins. After sCO₂ processing, the precipitate fraction was typically comprised of mainly α-LA and most of the minor whey proteins, with a non-negligible amount of co-precipitated or entrapped β-LG. On the other hand, gel electrophoresis results for the soluble fraction showed mostly β-LG, with a small amount of non-precipitated α-LA and little to no minor whey proteins. GPM did not appear on the gels.

The presence of large quantities of GMP in both the initial WPI and the soluble fraction after processing was identified with HPLC analysis, Figure 3. The retention times corresponding to the various proteins were determined using protein standards. The peaks between 3.7–4.0 min retention time were identified as GMP; the peaks for α-LA appeared between 17 and 19.5 min; the peaks corresponding to β-LG typically appeared between 22 and 24 min; and all the smaller peaks between 4.5 and 30 min were attributed to BSA, Ig, Lf or casein according to their standards.

Prior results found that the total amount of precipitated proteins and the relative proportions of all proteins in the two fractions greatly depend on the process parameters (time, C_WPI, T, and P) that influence the aggregation and precipitation kinetics of the different whey proteins, as described in detail by Bonnaillie and Tomasula [14]. For example, increasing the fractionation temperature, T, at constant pH and WPI concentration (i.e., constant P and C_WPI) considerably accelerated the precipitation rates of α-LA, β-LG, and the minor whey proteins out of the WPI solution. However, the individual aggregation rates and behaviors greatly depended on the values of T, C_WPI and pH. At all concentrations between pH 5.0 and 4.2, the precipitation of α-LA accelerated with T between 55 and 64 °C, and was maximized when T ≥ 64 °C, the thermal denaturation temperature of α-LA [46]. On the other hand, precipitation of β-LG increased slowly between 55 ≤ T ≤ 62 °C, then accelerated significantly when T ≥ 65 °C. Consequently, at all pressures between 8 and 31 MPa, sCO₂ treatment of WPI solutions between 60–62 °C produced moderate amounts of fairly pure α-LA precipitate, depending on pH, while processing at or above 65 °C produced larger amounts of precipitate with lower α-LA purity due to greater β-LG contents. The introduction of β-LG in the precipitate fraction was found to be two-fold: a portion of β-LG was entrapped within the α-LA aggregates due to water-holding of the β-LG-rich solution by the very hydrophilic α-LA aggregates [47], while another part of the β-LG precipitated due to thermal- and pH-driven denaturation and/or pressure or anti-solvent effects of CO₂. A study comparing the precipitation of β-LG with hydrochloric acid (HCl) and sCO₂ to differentiate pH and pressure/anti-solvent effects is underway in our laboratory.

WPI fractionation studies with HCl showed that the optimal combination of α-LA purity and yield in the precipitate fraction was obtained at pH 4.0–4.2, T = 60–62 °C, and a long residence time (t ≥ 1 hour). Experimental runs with the sCO₂ process were performed near the assumed range of optimal conditions: T = 60 to 65 °C, C_WPI = 5 or 10%, and P = 5.5, 8.3 or 31 MPa (corresponding to pH 5.0 to 4.4 depending on C_WPI). P = 8.3 and 31 MPa were chosen as low and high pressure values in the supercritical range, limited by the reactor’s maximum working pressure of 34 MPa. P = 5.5 MPa is the pressure of the liquid CO₂ tank at room temperature, and was added later as described below. Different combinations of parameters were used to vary recovery of the various proteins and produce fractions with different purities and yields, to analyze the relationship between processing costs and products quality and quantity. Table 1 presents the series of experimental conditions tested, with residence times of 2 hours for each run.

Table 1 shows that both the rates of precipitation of α-LA and β-LG are highly sensitive to temperature, WPI concentration and pH (C, P), which renders optimization of the sCO₂ fractionation process complex due to conflicting parametric effects on the purities and yields of the products, as well as on the economy of the process, as studied below.

In the range of processing conditions tested, purity of the solid α-LA fraction varied from ~39% to 61% (w/w), a two- to three-fold enrichment compared to the initial WPI, with α-LA recovery between ~60–98% in the solid fraction, and β-LG and GMP recoveries of ~63–89% and ~88–97% in the liquid fraction, respectively. The yield of solid fraction varied between ~21% and 40% of the initial WPI powder. To simultaneously compare the effect of processing parameters on both the precipitation rates of α-LA and β-LG, a ‘purity ratio’ was introduced, defined as the ratio of α-LA recovery over β-LG recovery in the solid fraction:

Between two sets of processing conditions, a larger value of the purity ratio signifies that the precipitation rate of α-LA was greater than that of β-LG.

At constant T and C_WPI, the total yield of precipitate generally increased with increasing sCO₂ pressure (i.e., decreasing pH), while the relative rate of α-LA/β-LG precipitation (purity ratio) decreased with P. This means that β-LG precipitation was more positively affected by P than α-LA precipitation, producing a greater solid yield with a lower α-LA purity, together with a lower β-LG recovery in the liquid fraction. This trend is opposite to that previously observed with HCl, where the rate of α-LA precipitation increased faster than that of β-LG when pH was lowered between pH 5.0 and 4.0 [14]. Thus, pressure or anti-solvent effects of sCO₂ may possibly be reducing the stability of β-LG in the CO₂-saturated WPI solution. According to these results, lower pressure experimental runs were added using the tank’s pressure of 5.5 MPa, with C_WPI = 10% and T = 60, 62 and 65 °C. Although out of the supercritical region, α-LA and β-LG followed the same precipitation trends, and α-LA fractions with a greater purity were obtained at all values of T.

At constant C_WPI and P, increasing T from 60 to 65 °C greatly increased the recovery of both α-LA and β-LG in the solid fraction in all cases, resulting in a much greater solid yield. However, the purity ratio was reduced considerably because the precipitation of β-LG accelerated faster than that of α-LA. From 60 to 62 °C (rows 1 and 6), the rate of α-LA precipitation increased noticeably, while β-LG precipitation was almost unchanged, resulting in both a higher purity ratio and a slightly higher solid yield.

Thirdly, reducing C_WPI at constant T and P had a null or negative effect on the recovery of β-LG in the solid fraction, mostly due to lesser β-LG entrapment by water-holding, and its effect on α-LA precipitation varied according to pH. The highest α-LA fraction purity, 61%, was obtained with C_WPI = 5% at 60 °C and P = 8.3 MPa (pH 4.7).

Because both the products quantity and quality were highly sensitive to the processing parameters, with conflicting trends, process optimization in terms of combined ‘yield & purity’ was not reached. The process was modeled using experimental results to calculate production costs as a function of the processing parameters, products yields, and purities, to identify feasibility and the most potentially profitable operating conditions.

Figure 4 presents a flow chart of a simplified continuous version of the sCO₂ whey protein fractionation process, including the following base equipment and material streams: a high-pressure stirred reactor to contact the WPI solution with sCO₂; a de-foaming tank to receive and break the foam that forms during depressurization to atmospheric pressure; a recycling loop to bring the CO₂ that escapes from the foam back to the reactor; a high-speed centrifuge to separate the α-LA particles from the β-LG-rich solution.

To calculate all the material flows, a medium-size process was modeled with a total production rate of 90.7 kg per hour (200 lb/h) for the dried fractions, corresponding to a feed solution (stream 1) of 90.7 kg/h WPI mixed with 816 kg/h water when C_WPI = 10%, or 90.7 kg/h WPI mixed with 1723.6 kg/h water when C_WPI = 5%. Material flow rates in streams 6, 7 and 8 were calculated with mass balances using the experimental individual protein recoveries derived from the SDS-PAGE and HPLC compositional analyses. Material flow rates in streams 2, 3, 4 and 5 were calculated using experimental data for the solubility of CO₂ in WPI solutions, and assuming complete vaporization, removal, and recycling of the dissolved CO₂. During extraction from the reactor and depressurization down to atmospheric pressure, dissolved CO₂ and whey proteins form a foam, which collapses in the de-foaming tank. We assumed full CO₂ recovery from the headspace of the de-foaming tank for recycling purposes. In reality, a small percentage of CO₂ remains solubilized after foam collapse, as evidenced by the difference between pH measured immediately after foam collapse, pH 6.0, and pH values up to 6.8 (depending on C_WPI) after the remaining CO₂ had time to diffuse out of the fractions (1–2 days). This non-recyclable portion of CO₂ was neglected in mass balance calculations. At operating conditions of 8.3 MPa and a 907 kg/h feed, an ideal CO₂ recycling loop would return 47 kg/h of CO₂ (~23,500 L/h at atmospheric pressure) to the reactor. Stream 2 serves only in the transient phase (initial CO₂ charge of reactor); once the process has reached steady-state, CO₂ circulates within the recycling loop only (streams 3, 4, 5).

Table 2 presents the material flow rates in streams 1 through 8 as calculated using the experimental data with C = 5%, T = 60ºC and P=31 MPa and assuming ideal CO₂ recycling. The partition between streams 7 and 8 in the centrifuge was calculated from the experimental recovery percentages of water and of each individual protein in the two fractions.

Mass balance calculations as presented in Table 2 were done at all values of T, P and C_WPI and all the calculated compositions (streams 7 and 8) matched the experimental values well and validated the accuracy of the SDS-PAGE and HPLC analyses.