Whole fish samples (sardines) were purchased from Pasar Borong, a whole sale local market at Pugung, Selangor, Malaysia. Chemicals and solvents were either of analytical or HPLC grade were purchased from Fisher Scientific Chemical Co. (Loughborough, England) and Merck (Darmastadt, Germany).

The fresh sardine samples purchased from Pasar Borong were kept in plastic bags and transported in an insulated icebox to the laboratory. The samples were immediately de-headed, gutted, and washed with copious amounts of cool water, and the flesh and bones were then separated using a de-boner (model- FD 6, Safe World Food-Tech Pvt. Ltd., Klang, Selangor Darul Ehsan, Malaysia). The flesh was stored at −25 °C and then freeze dried (Model: LABCONCO, USA) at a drying temperature of −40 °C under a 0.133 bar vacuum. The moisture content of the freeze dried samples was determined (data not given) and kept in desiccators until use.

Before oil extraction, the fish was analyzed for its proximate composition such as moisture, lipid, ash and protein contents to indicate its initial nutritional qualities. Moisture, oil and ash contents were determined as described by AOAC (AOAC, 1990) with slight modifications for Soxhlet oil extraction. The oven-drying method (105 °C) was used for moisture content and furnace at 550 °C for ash content. Protein content was determined according to Pomeranz and Meloan (Pomeranz and Meloan, 2000).

Soxhlet extractions were carried out (in triplicate) as described by AOAC, 1990 with minor modifications. Five grams of freeze dried fillet were extracted using 200 mL of petroleum ether for eight hours. The extracted lipids were evaporated under vacuum at 60 °C using a rotary evaporator (Rotavapor R-210, Büchi, Switzerland) and then were placed in an oven at 30 °C for 1 h before they were transferred into desiccators and reweighed. The extracted lipids were transferred into a brown bottle, flashed with nitrogen gas and stored at −25±1 °C.

Freeze dried sardine fillets were extracted using modified Kinsella (1977) method (MKM) by Kim et al. (1991). Initially, samples were homogenized for 2 min using a warring blender with chloroform and methanol in the proportion of 1 g tissue: 1 mL chloroform: 2 mL methanol. An additional equivalent amount of chloroform and de-ionized water (1 mL·g⁻¹ tissue) was added and the mixture was homogenized for 30 sec. The mixture was filtered through a Whatman No.1 filter paper on a No.3 Buchner funnel with a slight suction. The mixture was transferred to a decanter flask and was left to stand for a few minutes to complete the separation and clarification. The lower clear phase (chloroform and lipid) was poured into a conical flask. The extract was then concentrated using a vacuum rotary evaporator (Rotavapor R-210, Büchi, Switzerland). The extracted lipid was transferred into a brown bottle, flashed with nitrogen gas and stored at −25±1 °C until further analysis. The total lipid content was calculated gravimetrically as below:

Total lipid content (%) = W₁/W_S*100

Where: W₁ = weight of dried lipid (g) and W_S = the weight of sample (g).

The supercritical fluid extractor (SFE) used was ABRP200, Pittsburgh, PA, USA with a 500 mL extractor vessel attached (Figure 1). Parameters were selected from a software (ICE) program, which permitted to set and control the extraction status. The liquid CO₂ was pressurized to the desired pressure and heated to the targeted temperature with a pressure pump (P-50, Pittsburg, PA, USA) to reach the supercritical state prior to passing it into the extraction vessel. During this step, the pump head temperature was decreased to 4 °C. Then, the system was equilibrated until pressure, CO₂ rate and temperature became constant to begin the extraction. Absolute ethanol (EtOH) was used as the co-solvent and was fixed at a flow rate of 3 mL·min⁻¹.

The duration of the static and dynamic extraction times was fixed at 30 and 80 min, respectively. The thimble containing the sample (40 g) was placed in the extraction chamber equipped with a heating jacket. The components were then extracted by the pre-heated supercritical CO₂ and entered the trap through a nozzle where CO₂ was depressurized. The extraction was performed under various experimental conditions as generated by the RSM design. EtOH was removed from the extracts by vacuum evaporation using a rotary evaporator. The extracts were then placed in the oven at 30 °C for 30 min before being transferred into desiccators for final constant weight. The extracts were transferred into brown glass bottles, flashed with nitrogen and stored in a freezer of −25±1 °C until further analysis.

The fatty acid (FA) composition of the lipid extracts was analyzed based on the Christie (1993) method using Agilent gas chromatography (G1530N, USA). The column used was BPX-70 (60 m×0.32 mm i.d., 0.25 µm film thickness) with the phase composition 90% biscyanopropyl; 10% cyanopropyl phenyl polysiloxane from SGE, Melbourne, Australia. A100 µL aliquot of the test sample was thoroughly mixed by dissolving 50 µL of sample into 950 µl of n-hexane, and 50 µL of sodium methoxide was added to prepare the FA methyl esters (FAMEs). The mixture was then shaken vigorously using an auto-vortexer (Stuart, UK) for 30 s and stored for 5 min so that it formed two layers. The clear upper layer containing the FAMEs (1 µL) was pipetted off and injected into the gas chromatograph. The oven temperature was set at 115 °C, held for 2 min, raised to 180 °C at a rate of 8 °C·min⁻¹ and held for 10 min to be finally raised to 240 °C at a rate of 8 °C·min⁻¹ and held for 10 min and the carrier gas (helium) at a rate of 1.6 mL·min⁻¹ was flashed through. The FA components were identified based on the standard mixture of FAMEs containing 37 mixed fatty acids.

Response surface methodology (RSM) was used to determine the optimum conditions for yield EPA and DHA of sardine oil extracted using SFE. The experimental design and statistical analysis were carried out using the Minitab V. 14 statistical package (Minitab Inc., PA,USA). Central composite design (CCD) with cube points was selected to evaluate the effects of two independent variables (extraction temperature and pressure), coded as X₁ and X₂, respectively, on the yield EPA and DHA of the SC-CO₂ extracted sardine oil. The minimum and maximum values for the extraction temperature were set at 40 to 70 °C whereas pressure was in the range of 200 to 400 bars. It should be noted that using CCD with cube, out range values called “star points” could be found to predict the optimum point in case it lies out of the selected range. Optimization of the two independent variables was achieved by maximizing the three dependent variables i.e. yield, EPA and DHA to achieve highest values using a MINITAB numerical response optimizer. The whole design consisted of 13 combinations including five replicates of the center point as in Table 1, (Myers, 2002). The ANOVA tables were generated and the effect and regression coefficients of individual linear, quadratic and interaction terms were determined. The significances of all the terms in the polynomial were analyzed statistically by computing the F-value at a probability (p) of 0.001, 0.01 or 0.05 of the SC-CO₂ extracted sardine oil. The statistically found non-significant (p>0.05) terms were removed from the initial models and only significant (p<0.05) factors were involved in the final reduced model. Experimental data were fitted to the following second order polynomial model and regression coefficients were obtained. The generalized second-order polynomial model proposed for the response surface analysis was given as follows 1 y = β 0 + ∑ i = 1 k β i x i + ∑ i = 1 k β i i x i 2 X i 2 + ∑ i K - 1 ∑ j k β i j x i x j

Where β₀, βi, βii, βij were regression coefficients for intercept, linear, quadratic and interaction terms, respectively. X_I and Xj were coded values of the independent variables, while k was the number of the tested factors (k=2).

For the söxhlet and modified Kinsella methods, triplicate extractions of each were considered and their means were compared with the SC-CO₂ optimized sardine oil.

Moisture protein, ash and lipid contents are generally used as indicators of nutritional values of fish (Stansby, 1962). A high content of water in fish was correlated with low protein and lipid contents (Dempson et al., 2004). The moisture, protein, crude lipid and ash contents of the Malaysian sardine fillets were 77.8±0.2, 15.4±0.51, 5.1±0.10 and 1.4±0.17, respectively. These values were within the normal ranges for sardine which were reported by Payne et al. (1999), taking into account that lipid levels as well as other energy parameters fluctuate seasonally (Payne et al., 1999). Despite differences in region, which may lead to a variation in parameters (Çelik et al., 2005), our findings were in a close agreement with those reported by Serdaroğlu and Felekoğlu (2005), Payne et al. (1999), and Fernandes et al. (2014).

The ranges of each independent variable (pressure and temperature) which affected extraction efficiency of EPA and DHA ratios were chosen. In this study, the upper and lower values for the parameters were set at +alpha (+α=1.414) and −alpha (−α=1.414), hence, all the factor levels were selected within the limits that were practical with SC-CO₂ (above critical temperature of 31 °C and critical pressure of 72 bars) and desirable. In RSM, natural variables are transformed into coded variables that have been defined as dimensionless with mean zero and same standard deviation (Liyana-Pathirana and Shahidi, 2005).

The linear, quadratic and interaction effects of supercritical pressure and temperature on sardine oil extraction efficiency, and the ratio of EPA and DHA are shown in Table 1. The results suggest that the models fitted for response variables were empirically adequate due to their high coefficient of determination (R²>0.92), which means that more than 92% of the response variation could be explained as a function of the two SC-CO₂ parameters (Pressure and temperature). The highly adjusted R² (>0.89) as well as insignificance of any lack of fit in the data also indicated its reliability.

Using multiple regression analysis, the relationship between the tested parameters and the responses were explained in equations (Eq. 2, 3, and 4 for yield, EPA and DHA, respectively). 2 Yield =- 34.8 + 0.151 X 1 + 0.678 X 2 - 0.000185 X 1 2 - 0.00453 X 2 2 - 0.000572 X 1 X 2 3 EPA = 8.25 - 0.00346 X 1 - 0.0730 X 2 - 0.000009 X 1 2 + 0.000187 X 1 X 2 4 DHA = 14.8 + 0.0235 X 1 + 0.101 X 2 - 0.000031 X 1 2 - 0.00189 X 2 2

Multiple response optimizations were used to assess the optimum levels of the parameters which could achieve the desirable response areas (Mirhosseini et al., 2008). Besides the numerical optimization (Figure 2), the 3D plots (Figure 3) which were advocated for the graphical interpretation of the interaction effect of independent variables on the dependent variables (Montgomery and Wiley, 2001) were also considered using the Minitab software to locate the exact optimum point of independent variables and to obtain overall joint optimized values using the Minitab program response optimizer. The overall optimal values for maximum SC-CO₂ extraction efficiency and n-3PUFAs ratios (EPA and DHA) of pressure and temperature were anticipated to be at pressure of 328 bars and temperature of 40 °C as shown in Figure 2. Under the optimized conditions, the corresponding predicted dependent variables for yield, EPA and DHA were 7.20, 5.68 and 20.09%, respectively.

The fitness of the response surface equation was checked from the error rate between experimental and predicted values of the reduced response regression models (Tables 2 and 3). The experimental and predicted values of yield, EPA and DHA obtained from equations 2, 3 and 4 are presented in Table 1. For each of the experimental values, Y₀ was compared with the predicted values, Y₁ calculated from the equation. A close agreement between the experimental and predicted values was noted and no significant (p>0.05) difference was found between those values, thus suggesting the adequate fitness of the response equations.

The sardine oils obtained from the various SC-CO₂ (run orders 1–13) and solvent (Soxhlet or MKM) extracts are given in Table (1). Although some SC-CO₂ runs (center points) gave higher extraction efficiency than MKM, overall results of both were statistically similar (p>0.05) and the two techniques gave a higher yield (>6%) than that of the söxhlet extraction (5.1%). SC-CO₂ extraction results were in agreement with those reported by Létisse et al. (2006). The MKM method initially established by Bligh and Dyer (1959) could be extracted with all lipids, including polar lipids, phospholipids and possibly lipids bound with other components from cellular membranes and with high yield. Yet application of this method for food supplementation has raised questions on its safety owing to toxicity of the solvents used (Létisse et al., 2006). Hence, the SC-CO₂ method using the environmentally-friendly and non-toxic CO₂ with small milliliters of non-hazardous solvents like ethanol or even nil could be a more attractive technique, provided that the conditions whereby maximum yield of lipid and/or fatty acids can be achieved, are optimized.

The individual influence and efficient interaction between pressure and temperature were investigated. The impact of pressure and temperature on the SC-CO₂ sardine oil extraction was determined at pressures of 200 and 400 bars, temperatures of 40 and 70 °C respectively, and at a constant CO₂ flow rate (18 g·min⁻¹). Similar trends of pressure and temperature influence on the extracts with SC-CO₂ were observed in previous studies (Mariod et al., 2010, Jamilah et al., 2011). The high extraction yield (>8%) was obtained at 300 bars and 55 °C followed by 200 bars and 70 °C, while the lower yield was obtained at 158 bars and 55 °C. Based on these results a significant (p<0.05) effect of pressure and temperature on the sardine extraction efficiency was obvious. At low and extremely high pressures (200>p>350 bars), the oil extraction efficiency was decreased. The consequence of the extraction pressure and temperature on the yield of sardine oil at a steady CO₂ flow rate (18 g·min⁻¹) is demonstrated in Figure 3a. In view of the above results, it is clear that there was a significant joint (p<0.05) effect between the two parameters; i.e., when low pressure (200 bars) but high temperature (70 °C) and vice versa (400 bars and 40 °C) was applied, a good yield of 6.21% and 6.04, respectively (Table 1) was achieved. A reciprocal impact of SFE pressure-temperature inter-relationship, similar to the present findings, was reported by Wie et al. (2009). By increasing the temperature or pressure of the solvent, the rate of extraction with SC-CO₂ can be improved (Zaidul et al., 2007). The center points (300 bars and 55 °C), however, showed the highest yields of sardine oil extraction with SC-CO₂. This was in line with what Pan et al. (2012) reported.

Despite the fact that the comparison of fatty acid composition in fish oils is quite difficult due to several probable affecting factors like season (Celik, 2008, Rasoarahona et al., 2005, Shirai et al., 2002) the EPA and DHA overall values extracted with MKM or SC-CO₂ in this study were in good agreement with those of South east Asian waters sardines (Chaijan et al., 2006). A significant (p<0.05) effect of pressure and temperature, with the latter being more influential on EPA and DHA, was noted. There was an inverse relationship of temperature increment with EPA (Figure 3b) and DHA (Figure 3c). For instance, when an extraction temperature as low as 33 °C with an intermediate pressure of 300 bars (Table 1) was used, the highest values of EPA (5.84) and DHA (20.35) were achieved. In contrast, the lowest portions were reached with high temperatures of up to 70 °C and a low pressure at 200 bars (Table 1). The severe susceptibility of poly unsaturated fatty acids to higher temperatures or light is a well addressed concern.

Lipids from freeze-dried sardine extracted through various extraction methods (Soxhlet, MKM and SC-CO₂) were analyzed in order to determine the relationship between changes in the lipid fatty acid (FA) profile and method of extraction. The extraction efficiency (%) of the sardine oil obtained from SC-CO₂ was varied based on the extraction conditions (Table 1), therefore, triplicate analyses of SC-CO₂ optimum extraction efficiency (pressure: 321 bars and temperature 54 °C with a corresponding yield of 8.04%) were compared with the conventional extraction methods. The percentage values of methyl ester FA analyses in the söxhlet, MKM and SC-CO₂ extracted oils are presented in Table (4), while their typical chromatographic peaks are shown in Figure 4. No significant difference (p>0.05) in FA composition between MKM and SC-CO₂ was apparent. However, both formers were significantly (p<0.05) different from the soxhlet extracted oil FA composition. The prolonged extraction time (8 h) at a relatively high temperature (60 °C) with Soxhlet had perhaps decreased the heat sensitive unsaturated fatty acids, especially the more susceptible PUFAs such as, EPA and DHA in the oil. From the analysis, the main components of sardine oil are shown in Table 4.