The full experimental design and corresponding data obtained are shown in Table 1, where the highest yield was obtained at run order 10, followed by 29, 19, 13 and 30 (Table 1). The combined effect of temperature, pressure, flow rate and soaking time compensated gaining the density of supercritical CO₂, which meant increased solubility of the solute in the solvent. Temelli et al. [30,31] reported that the impact of temperature is dependent on competing parameters; the CO₂ density decreases with temperature while the vapor pressure of the solutes increases, leading to the well-known crossover phenomenon for solubility isotherms. Thus, the resultant impact of temperature on solubility is dictated by whichever parameter is greater at a given pressure. In addition, diffusivity increases with temperature enhancing mass transfer kinetics during extraction [30–32]. The statements are in agreement with our observation of the effect of pressure, temperature and flow rate on the recovery of the total oil yield, as shown in Table 1. In general, extraction yield increases with pressure, due to an increase in lipid solubility in supercritical CO₂, and based on an increase in CO₂ density. Thus, the highest yield (67.0%) was obtained for the combined effect of 40 MPa, 57.5 °C, 2 mL/min and a soaking time at 2.5 h. This yield was reasonable to the yield (78.0 g/100 g sample, on dry basis) extracted using Soxhlet method.

In order to determine accuracy, the coefficient of determination (R²) was calculated for each run. For a good fit model, R² should be at least 0.80 [33], and in our significant model it was found to be 0.996. The regression coefficient values of the corresponding variables for this experiment are presented in equation 1. The oil yield (Y) was estimated by the second order polynomial equation shown below:

Where, Y represents the oil yield of applied temperature (X₁), pressure (X₂), flow rate (X₃) and soaking time (X₄). From Equation 1, it was observed that the linear terms: temperature, pressure and soaking time had negative effects whereas flow rate had a positive effect on oil yield. However, all of the independent variables e.g., temperature, pressure flow rate and soaking time, were the most significant parameters in SC-CO₂ extraction. Two quadratic terms (X₂² and X₄²) had positive and two (X₁² and X₃²) had negative effects on the yield. On the other hand, the interaction terms, X₂*X₃, X₁*X₂, X₁X₃, X₁X₄ and X₂*X₄ had positive effects whereas X₃X₄ had a negative effect on the yield. However, the optimum conditions were obtained at 57.5 °C, 40.0 MPa, 2 mL/min and 2.5 h for temperature, pressure, flow rate and soaking time, respectively, with the maximum oil yield of 67.0%.

For a better understanding and representing of the significant (p < 0.05) statistical interaction of factors in response, a three dimensional (3D) surface data plot is highly recommended by Xu et al. [34], which is shown in Figure 1. It was sketched by keeping the temperature and pressure constants low, following which the oil yield was also found to be low. The yield significantly increased at relatively high pressure (>28 MPa) and medium to high temperature from 55 °C to 65 °C. The increasing effect of temperature and pressure was statistically significant (p < 0.05) on oil yield up to 65 °C. Above 65 °C, the oil yield gradually started to decrease with pressure, which is probably due to the reduction of CO₂ density and solvation power of the SC-CO₂. However, the effect of pressure changes (10 MPa to 40 MPa) on oil yield had a more noticeable and significant effect than that of temperature. Pressure elevation at a given temperature enhances the solubility of the oil in SC-CO₂, which might improve the total oil yield [35]. Similar trends were also observed in the extraction of silkworm pupal oil using SC-CO₂ [36,37]. Authors concluded that temperature has a negative effect on yield after a certain limit.

According to Figures 1 and 2, pressure and flow rate had similar effects on oil yield at constant temperature, thus the temperature interacted with pressure and flow rate. The solvent density, vapor pressure, and the solubility mode of the solute depend on these two variables. Consequently, it is very difficult to predict and elucidate the effect of temperature on oil yield. At constant pressure, the oil yield increased significantly with both temperature and flow rate (Figure 2), since high temperature decreases the density of CO₂ and thus leads to an increase in vapor pressure of the solutes that improves the solubility as well as the mass transfer rate of solute, resulting in easier extraction of the desired compounds [25,36,37].

Figure 3 shows a significant interaction between pressure and flow rate on oil yield. Raising the pressure at constant temperature increased the density of CO₂ as well as the extraction efficiency of the solutes. However, performing the extraction using SC-CO₂ at high pressure is not always advisable, as at high pressure levels where the highly compressed CO₂ may provoke a complexity in the extraction by introducing a repulse reaction between the solute and solvent [38]. Therefore, the combined effect of pressure and flow rate was not as significant as shown in Figure 3, while the flow rate was interacting with the temperature. Moreover, this leads to a decrease in yield at the time of fluid decompression, either by increasing analyte loss or by using an elevated pressure drop through the extraction cell [38].

The maximum oil yield (67.0%) was obtained from viscera at optimum conditions. Our results are similar to the results of Rubio-Rodríguez et al. [39], who found 63% hake oil from offcuts using supercritical fluid extraction. In another study, about 53.2% of oil was reported from the skin of Indian mackerel in using various techniques of supercritical fluid extraction [23].

Fresh African Catfish (Clarias gariepinus) samples were collected from a local market in Malaysia. A cylinder of carbon dioxide with a purity of 99.99% was purchased from Malaysian Oxygen Ltd. Kuala Lumpur, Malaysia and all other solvents and chemicals used in this experiment were analytical grade and obtained in Malaysia.

The samples were immediately de-headed and washed with copious amounts of fresh water to separate the viscera. The viscera were then stored overnight in a freezer at −18 °C, and then freeze dried (Model: LABCONCO, USA) at a constant drying temperature of −47 °C and vacuumed at 0.133 bar. The dried samples were ground using a blender and stored in an airtight glass bottle in a coldroom at 6 °C pending laboratory use.

The moisture content of the dried sample was determined by the oven dry method [40]. Five grams of finely ground sample was place in pre-weighted ceramic crucibles before being put into the oven. The temperature of the oven was set at 105 °C, and the heating process continued until constant weights of the samples were achieved. Then, the crucibles were transferred to a desecrator to cool before reweighing. The difference of the two weights (initial and final) indicated the moisture content and it was found to be 3.95% in viscera.

Total lipid content of the catfish viscera was determined by Soxhlet extraction method. Five grams of finely ground dried samples were extracted with 200 mL of petroleum ether with three replications over an extraction period of 8 h. Extra water and petroleum ether residue in the extracted oils were evaporated using a rotary evaporator (Heidolph, Germany) at a temperature of 45 °C. The evaporated sample was then dried in an oven at 45 °C for 1 h. Total lipid content of the catfish viscera was 78 ± 0.6% based on the dry weight of the sample.

All the runs were carried out in a supercritical fluid apparatus (Model PU-1580, Jasco Corporation, Tokyo). For each trial, 5 g of dry sample was loaded into a 10 mL extraction vessel (model Ev-3, Jasco Corporation, Tokyo), and then placed into an external water bath at a temperature ranging from 35 °C to 80 °C. Then, the valve of the CO₂ cylinder was opened and the CO₂ allowed to circulate through the cooling jacket of the chiller to cool before reaching the extraction vessel at a constant flow rate ranging from 1 mL/min to 3 mL/min (Model 631 D, Tech-Lab Manufacturing sdn. Bhd., Selangor, Malaysia), where CO₂ gas was converted to liquid form. After reaching the desired pressure in the extraction vessel, the CO₂ valve was closed for a certain period of time to soak the sample in pure CO₂: regarded as the “soaking time” for this experiment. A back pressure regulator (BPR) (model BP-1580-81, V, Jasco Corporation, Tokyo) was used to control the system pressure and separate the CO₂ from the extract. Then, the CO₂ valve was opened again during continuous extraction at constant pressure, temperature and flow rate. At each condition, experiments were conducted in duplicate, and each yield was the mean of duplicate measurements. Finally, the yield trap was collected and stored at −18 °C for further analysis.

A central composite design consisting of 30 experimental runs with six replications at the central points were employed to optimize the extraction variables, namely temperature, pressure, flow rate and soaking time. The creation of design matrix, experimental data analysis and optimization were all undertaken using Minitab v.14 statistical software. The polynomial equation represents all possible combinations of the extracting variables (X₁, X₂, X₃ and X₄) of their main, quadratic as well as the interacting effects on the response variable of oil yield (Y). A preliminary study was conducted to select the range values of the parameters (Table 2). More emphasis was given for the selection of temperature level: the lower limit was 35 °C, just above the critical points of CO₂ (31.1 °C) and the upper limit was not more than 80 °C to save the thermally sensitive compounds from thermal degradation [41]. All the design points were performed three times except the centre point. Experiments were run in randomized order to minimize the effect of unexplained variability induced by extraneous factors. The polynomial regression equation presented below was used for predicting the response variable (Y).

Where, Y is the response (percentage of oil yield) β₀ is a constant and β_i, β_ii, β_ij are the linear, quadratic and interaction terms, respectively. X_i and X_j are the levels of independent variables.