Wheat germ was supplied by a local wheat processing industry, Andhra Pradesh, India. The raw wheat germ was pretreated by heating in an oven at 105 ^°C for 30 to 45 min to remove the possible free moisture and to inactivate the enzymes. Then the sample was kept in tightly sealed plastic bags for further extraction. All the chemicals and solvents used in this study were purchased from SD Fine Chemical Ltd. (Mumbai, India) and were of laboratory reagent grade.

The oil in the wheat germ was extracted using a SC-CO₂ extraction unit with 500 cm³ capacity supplied by the Waters Corporation, Milford, USA. The unit contains a chiller, CO₂ pump, co-solvent pump, heat exchanger, extraction vessel, automatic back pressure regulator, fraction collector and other accessories. The system is designed to withstand pressures and temperatures up to 60 M Pa and 80 °C, respectively. The system is controlled by a programmable logic controller (PLC). Initially, 150 g of wheat germ sample were taken in a 0.45 μm cotton filter bag and inserted into the extraction vessel and the chiller temperature was brought down to 5 °C. The CO₂ filled cylinder was connected to the unit and CO₂ was allowed to flow through the system. The CO₂ gas was pre-liquefied by passing through a shell side heat exchanger and then pumped into the extraction vessel where the temperature and pressure were kept at the desired conditions. The supercritical state of CO₂ was achieved at this stage. The oil extraction was carried out and the extract was collected in a sample bottle from the collection vessel and the collected oil was dried and weighed as yield. The sample bottle was tightly sealed and kept in a refrigerator for further analysis.

The oil was also extracted from wheat germ in a soxhlet apparatus with hexane and the extraction procedure was continued up to 8 h to extract the maximum amount of oil. The oil content was determined as a percentage of the extracted oil of the sample weight (w·w^-1).

The Box Behnken design from response surface methodology (RSM) was used to perform the SC-CO₂ extraction. Three factors were included in the model: Pressure (X₁); Temperature (X₂); CO₂ flow rate (X₃). The coded and uncoded levels of both independent variables and experimental design are provided in Table 1 and low, middle and high levels are coded as -1, 0 and +1, respectively. With the center point assigned as ‘5’, 17 runs in total were obtained using the Box Behnken design.

The responses assessed using the Box Behnken design were oil yield (Y₁), phosphorous content (Y₂) and tocol contents (Y₃). These contents were estimated after each experimental run. The SC-CO₂ extraction time was fixed at 3 hrs, which was based on a few initial experimental trials. The responses (Y) investigated were expressed in second order polynomial equation (Eq. 1) as a function of independent coded variables (X₁, X₂,.…, X_n) affecting the responses. The regression coefficients for intercept, linear, quadratic, are β₀, β_i (i = 1, 2,.., k), β_ii (i = 1, 2, …, k), and β_ij (i = 1, 2, …., k, j = 1, 2, …., k), where k is the number of variables.

A statistical analysis was performed using RSM software Design Expert^® v.10. The responses were analyzed using analysis of variance (ANOVA). The efficiency of model was ascertained by determined coefficient (R²) and its implication was found by an F-test.

The tocols present in the wheat germ oil were analyzed according to the AOCS prescribed analytical method (AOCS 1994 Ce 8–89) using high performance liquid chromatography (HPLC). Commercially available α-Tocopherol (> 96% purity), β-Tocopherol (> 90% purity), γ-Tocopherol (> 96% purity) and δ-Tocopherol (> 90% purity), Sigma Aldrich, USA were used as standards. An Agilent 1100 series HPLC unit equipped with a fluorescence detector was used. The normal phase silica column (LiChroshper Si-60; 250 mm × 4.0 mm × 5 μm) (Merck Millipore, UK) was used. The excitation and emission wave lengths of the detector were maintained at 292 and 330 nm, respectively. The isocratic mobile phase containing hexane and isopropyl alcohol (99:1, vol·vol^-1) was used at a flow rate of 1.0 cm³·min^-1. The total tocol (vitamin E) content was measured in parts per million (mg·Kg^-1).

The phosphorous content in the samples was estimated using the International Union for Pure and Applied Chemistry (IUPAC) method (Pacquot and Hautfenne, 1987). A sample (2-10 g) was taken in a 50-100 cm³ silica crucible along with magnesium oxide (0.1 g) burnt in a fume hood. The crucible was put in a muffle furnace at 800 ^oC for 1 hr to obtain the ash. This ash was then dissolved in exactly 5 cm³ of an aqueous nitric acid (6 N) solution. 20 cm³ of 1:1 aqueous ammonium molybdate and ammonium vanadate solutions were added to the sample. A blank solution was also prepared under the same conditions. The absorbance was recorded at 460 nm using UV-spectrometer (Perkin Elmer, Lambda 35) for both sample and blank and the corresponding concentrations were taken. Phosphorous content was calculated accordingly as per the method.

The SC-CO₂ extraction of wheat germ was carried out under 17 experimental conditions which were obtained from the Box Behnken design. All experiments were carried out in triplicate at a fixed extraction time of 3 h by varying pressure (30-50 M Pa), temperature (40-60 ^°C) and CO₂ flow rate (10-30 g·min^-1). The oils obtained for each experiment were analyzed for phosphorous content and tocol contents. The Box Behnken experimental design and observed responses are shown in Table 2. The surface plots of the yield, phosphorus content and tocol contents as a function of the various independent variables such as pressure, temperature and CO₂ flow rate are shown in Figures 1-3. Keeping the third variable constant, the response surface plots were obtained by varying two independent variables. The significance of linear, quadratic and interaction coefficients were studied at the p values of less than 0.0001, 0.005 and 0.05. The oil yield, phosphorous content and tocol content values varied in the range of 3.92 to 9.8%, 26 to 48.4 mg·Kg^-1 and 1384 to 2286.3 mg·Kg^-1, respectively, in the Box Behnken design matrix. It was found that the effect of pressure was more eminent on the extraction yield compared to other parameters. As expected, the oil yield increased with increasing pressure as increase in pressure leads to an increase in the density of SC-CO₂ thereby increasing the solubility of the oil in SC-CO₂ (Ghoreishi et al., 2016). As evident in Table 2, the high pressure of 50 M Pa yielded more oil compared to the low pressure of 30 MPa. On the other hand, the rise in temperature decreases the density of SC-CO₂ and solubility decreases, due to which the oil yield decreased with the increase in temperature (Zacchi et al., 2006; Kamali et al., 2015). Therefore, the maximum extraction oil yield was observed at the low temperature of 40 °C. The increase in CO₂ flow rate resulted in a slight increment in oil yield. The effects of processing parameters on the yield are shown in the response surface plots in Figure 1.

The phosphorous content initially increased with the increase in pressure and temperature at a constant CO₂ flow rate and then was decreased gradually. A small increment in phosphorous content was also observed with increasing CO₂ flow rate. Figure 2 shows the effects of process parameters on phosphorus content in the SC-CO₂ extracted oils. The tocols contents in the oil decreased with an increase in pressure and increased when the temperature was raised from 40 to 60 ^°C. A small increment in tocol contents was observed with increasing CO₂ flow rate. Figure 3 shows the dependence of tocol contents on the pressure, temperature and CO₂ flow rate.

The coefficients of the second order polynomial equations derived from the models are shown in Table 3. The linear coefficients of X₁, X₂ and X₃, and the quadratic coefficient of X₁² for oil yield, the linear coefficients of X₁ and X₃ and the quadratic coefficients of X₁² and X₂² for phosphorus content, the linear coefficients of X₁, X₂ and X₃ and the quadratic coefficients of X₁² and X₂² for tocol content are significant at p < 0.0001. Similarly, the quadratic coefficient of X₃² for oil yield, linear coefficient of X₂, the quadratic coefficient of X₃² and the interaction coefficient of X₁ X₂ for phosphorus content, the quadratic coefficient of X₃² and the interaction coefficient of X₂ X₃ for tocol contents are significant at p < 0.005. The interaction coefficients of X₁ X₂ and X₁ X₃ for oil yield, and the interaction coefficient of X₁ X₃ for phosphorus content are significant at p < 0.05, whereas the other coefficients are insignificant.

R-square was estimated to study the sustainability of the second-order polynomial equations and models (Daneshvand et al., 2012; Banik and Pandey, 2008). For a good fit of a model, the R-square value should be at least 0.80 (Joglekar and May, 1987). The R-square values for the oil yield, phosphorous content and tocol contents were found to be 0.9961, 0.9970 and 0.9956, demonstrating a good correlation between the experimental data to the predicted data by the three models. Variations of 0.39%, 0.3% and 0.44% were observed for the responses predicted by the models for oil yield, phosphorous content and tocol contents, respectively, but they cannot be explained by the model. The predicted R-squared values (0.9722, 0.9609 and 0.9519 for the oil yield, phosphorous content and tocol contents) are in good agreement with the adjusted R-square values (0.9912, 0.9932 and 0.9899 for the oil yield, phosphorous content and tocol content) and the values were within very narrow differences (less than 0.2) for each other. Table 4 shows the F value with the corresponding p value of models, residual and lack of fit for oil yield, phosphorous content and tocol contents. The p values for lack of fitness were found to be more than 0.05 for all three responses and that indicates that lack of fit was insignificant and that the three models fit the data well.

To attain the optimum conditions and to predict the yields, three second order polynomial equations for oil yield, phosphorous content and tocol contents were used. In the Box Behnken design, a desirability function was established which will simultaneously satisfy the requirements of all the responses. Using the desirability function, it is possible to get a point by numerical optimization that will maximize or minimize the function. By adjusting the weight or importance, the goal was set which was combined into one desirability function for multiple responses. The goal for extraction parameters in the SC-CO₂ extraction was set in range where the goal for oil yield and tocol contents were defined as maximum and phosphorous content was defined as minimum. The weight factor which illustrates the shape of the desirability function of the three responses was fixed as 1 and the “importance” value of 3 was chosen for all the goals in the study.

The optimum extraction conditions for SC-CO₂ extraction obtained by the application of RSM were found to be pressure of 50 M Pa, extraction temperature of 60 ^°C, CO₂ flow rate of 30 g·min^-1. The optimum conditions resulted in the predicted oil yield of 8.87%, phosphorous content of 31.86 mg·Kg^-1 and tocol content of 2059.92 mg·Kg^-1 with a desirability factor of 0.775. To validate the sustainability of the models and the predicted values obtained, an experimental run was carried out under the optimal conditions and the results obtained were 8.68%, 42.5 mg·Kg^-1 and 2057.3 mg·Kg^-1 for oil yield, phosphorous content and tocol contents respectively. The experimental results were found to be within the range of predicted values, indicating that the optimization of SC-CO₂ extraction using the Box Behnken design was adequate to explain the process. Conventional solvent extraction using hexane as solvent resulted in 9.4% oil yield, 1056 mg·Kg^-1 phosphorous content and 2046.4 mg·Kg^-1 tocol content. Although the yield is shown to be a bit less, this may be due to the poor extractability of phospholipids by SC-CO₂.