3.1. MAE Process Optimization
The influence of the independent variables time, microwave power, and solid/liquid ratio on the MAE of oil from fish by-products was studied using RSM. The oil yield results obtained over the 20 runs of the experimental design from the two samples of fish by-products (S1 and S2) are shown in
Table 2. The extraction yield varied with a function of the applied MAE conditions and ranged from 12.57 to 18.26 g/100 g dw in S1 and from 14.88 to 20.99 g/100 g dw in S2. Although the oil yields were higher from S2, the lowest and highest yields achieved with both fish by-product samples were obtained from the 5th and 13th runs of the experimental design, respectively. On the other hand, the GC-FID analysis showed that the fatty acid profile did not change significantly regardless of the applied MAE conditions (data not shown). Therefore, the extraction process was optimized when considering only the oil content as a dependent variable.
The experimental data in
Table 2 were fitted to the second-order polynomial Equation (2) to construct the theoretical model demonstrated in Equations (6) and (7), which were used to predict the optimal MAE conditions that maximize oil recovery from the fish by-products S1 and S2, respectively.
Equations (6) and (7) show that the three independent variables (t, P, and R) significantly affected (p < 0.05) the fish by-product oil extraction. The higher model coefficients translate to a more marked linear, quadratic, or interaction effect, regardless of its sign. Furthermore, the existence of interactions at least between two independent variables justified the use of RSM as an optimization tool, as one-factor-at-a-time approaches are unable to evaluate these terms and involve a greater number of tests, making them more costly and time-consuming.
The statistical data of the model fitting procedure are presented in
Table 3, where one may observe that both models are statistically significant (
p < 0.05) and present
R2 and
R²
adj values greater than 0.95 and 0.93, respectively, thus indicating that the variability is explained by the independent variables.
R²
adj is lower than
R2 because it penalizes the inclusion of terms that do not contribute to explaining the variability, and it is preferable whenever there are many terms in the model. This agreement between predicted and actual values is visually represented in the statistics diagnostic sections of
Figure 2 and
Figure 3, where the residual distribution is also shown to describe the behavior within the runs and show that there is no type of pattern that can influence the modeling. In addition, adequate precision values above 22.3 evidenced adequate model discrimination and the relatively low (≤2.7) coefficient of variation or relative standard deviation denoted a high degree of accuracy and reliability. A non-significant lack-of-fit (
p > 0.05) also characterized both models, indicating that they fit the data in
Table 2 well. Therefore, both models were statistically valid to navigate the design space and predict the optimal MAE conditions.
The 3D surface plots shown in
Figure 2 and
Figure 3 were constructed to visually represent the effects of the process variables on the oil content obtained from S1 and S2, respectively. For each 3D plot, the excluded variable was fixed at its optimal value. The extraction trends translated by the coefficients of Equation (6) and represented in
Figure 2 for S1 showed the marked linear (−1.50) but also quadratic (0.45) effects of the solid/liquid ratio. Increases in this independent variable led to lower oil yields (since the
b3 coefficient was negative). It also interacted with time (−0.3) and microwave power (0.37). Thus, the yield was promoted by the combination of low solid/liquid ratios with long irradiation times and low microwave powers. The extraction time also induced positive linear effects (0.78). Although the linear term coefficient (0.04) of microwave power was not significant (
p > 0.05), it was necessary for the model hierarchy. In turn, Equation (7), which referred to the extraction of oil from the fish by-product S2, was more marked by quadratic terms (
b11 = −1.2 and
b22 = −1.0) than Equation (6), which are represented by the response surfaces in
Figure 3 and by the individual 2D responses in
Figure 4. The microwave power induced the strongest linear effects (1.2), followed by solid/liquid ratio (−1.0), and then time (0.52). An interaction effect (−0.3) between microwave power and solid/liquid ratio also characterized Equation (7).
In order to determine the optimal MAE conditions to maximize oil recovery from each fish by-product, the three independent variables were set within the experimental range and the response variable was “maximized.” The model Equations (6) and (7) estimated obtaining 18.6 ± 0.3 g/100 g dw and 22.0 ± 0.4 g/100 g dw of fish oil from S1 and S2, respectively, when processed at 455.5 W for 20.6 min at 72.1 g/L or at 750.2 W for 17.3 min at 70.0 g/L, respectively. The optimal conditions for time and solid/liquid ratio were similar for both samples, while the microwave power was the independent variable that most diverged. This deviation can be deduced from the 2D responses of each independent variable represented in
Figure 2. As such, and since the batches of fish by-products generated by the industry can present certain compositional differences (e.g., variable proportions of different fish waste), global MAE conditions were determined considering the experimental data obtained with both samples, S1 and S2, simultaneously. This second optimization step showed that 18.3 ± 0.3 g/100 g dw and 21.4 ± 0.3 g/100 g dw of fish oil can be obtained from S1 and S2, respectively, when processed at 585.9 W for 18.7 min at a solid/liquid ratio of 70.9 g/L. Still, the oil yields obtained using these global conditions are similar to those predicted with the individual ones determined for each sample.
Some of the observed extraction trends can be compared with those described by de la Fuente et al. [
16] for the MAE of oils from salmon side streams, namely higher oil yields when using lower solid/liquid ratios (~80 g/L for backbones and heads); medium processing times (~11 min for heads and ~14.5 min for backbones and viscera); and low or high microwave powers (50 W for heads and 961 W for viscera, though this variable was not significant for backbones). These results also show a greater discrepancy for the ultrasonic power, which corroborates our observations and may be justified by the variety and different intrinsic nature of the fish tissues that make up the by-products and their interaction with microwaves. In another work, Rahimi et al. [
23] investigated the effects of MAE time on oil recovery from a homogenized mixture of fresh sardine heads, tails, and bones. Oil yields of 3.3 to 8.1 g/100 g fresh weight (fw) were obtained by increasing the irradiation time from 2 to 10 min using water as a solvent, while yields of 6.1 g/100 g fw were achieved by conducting irradiation over 4 min with a hexane:isopropanol ratio of 3:2
v/
v. However, these were the maximum times tested by the authors, and extraction trends show that higher yields could be reached by processing longer. Still, the methods used by Soxhlet and by Hara and Radin yielded only 58% and 20% when compared to MAE.
3.2. Extraction Yields of Fish By-Product Oils Obtained through MAE and SE
To validate the predictive ability of Equations (6) and (7), the individual optimal MAE conditions discussed above were applied experimentally to obtain fish by-product oils. As shown in
Table 4, 17.9 ± 0.8 g/100 g dw and 20.6 ± 0.9 g/100 g dw of fish by-product oil yields were obtained from S1 and S2, respectively, values which are in agreement with the model-predicted ones (18.6 ± 0.3 g/100 g dw and 22.0 ± 0.4 g/100 g dw), as confirmed by the post-analysis performed with the Design-Expert software. When comparing the S1 oil yields obtained through SE (18 ± 1 g/100 g dw) and optimized MAE, no statistically significant (
p < 0.05) differences were found. However, for S2, the MAE process was only able to obtain 61% of the oil recovered from this sample using SE (34 ± 1 g/100 g dw). Even so, the shorter extraction times (17–21 times shorter) and lower solvent consumption (5 times less) associated with this non-conventional method stand out compared to SE.
The efficacy of MAE to extract fish oil was recently investigated using sea bream (
Sparus aurata) and sea bass (
Dicentrarchus labrax) heads [
21]. The oil yield results (20.8–21.5 g/100 g dw), which corresponded to 52–55% of the total oil content, were similar to those of the present work. However, they were lower than those described by de la Fuente et al. [
16] for salmon side streams (38–77 g/100 g dw). For instance, the best results were observed in salmon viscera, which reached 77 g/100 g (SE) and 71 g/100 g (optimized MAE), thus allowing 92% of the total oil to be recovered. For salmon backbones and heads, the oil recovered under optimized conditions was 69%. These MAE processes took up to 33 times less time and 5 times less solvent than SE. The potential impact of the matrix is thus somewhat observed. According to the microscopic observations made by Costa and Bragagnolo [
24], MAE causes the rupture of fish tissue and consequent faster mass transfer, which explain the shorter extraction time required. In addition, MAE can reduce the extraction time by up to 90% and the solvent consumption by up to 25% compared to the conventional Folch method.
3.3. Fatty Acid Profile of the Fish By-Product Oils Obtained through MAE and SE
The fatty acid profile of the fish oil by-products obtained through SE and MAE consisted mainly of monounsaturated fatty acids (MUFAs, ~51%), given the high relative percentages of oleic acid (C18:1
n9) (
Table 4), an odorless olive oil associated with multiple beneficial effects on human health [
25]. Palmitic (C16:0) and docosahexaenoic (DHA, C22:6
n3) acids ranked second, with contents reaching ~14% (in S2 oil) and ~13% (in S1 oil), respectively. Relative percentages of linoleic (C18:2
n6) and eicosapentaenoic (EPA, C18:2
n6) acids up to 10 and 7%, respectively, and of α-linolenic acid (C18:3
n3) up to 4.9% were also found in the studied fish oils. These last four fatty acids were the ones that most contributed to the 28–30% of polyunsaturated fatty acids (PUFAs). As shown in
Table 4, neither the relative percentages nor the classes of fatty acids varied significantly (
p > 0.05) with the extraction method. In general, the fatty acid composition of the studied oils was similar to that of oils from sea bass, sea bream, and salmon heads, as well as salmon backbones and viscera, all obtained through MAE and SE, in which oleic acid also stands as the most abundant (~34–39%), followed by linoleic (13–18%), palmitic (10–15%), docosahexaenoic (DHA, 7–14%), and eicosapentaenoic (EPA, 4–6%) acids [
16,
26]. These by-products thus had higher linoleic acid percentages than the by-products used in the present study. Oleic acid was also the main fatty acid in Atlantic salmon head, frame, and viscera oils (38.8–44.2%) obtained using enzymatic hydrolysis [
11], and in cephalopod (
Sepioteuthis lessoniana) liver viscera oil extracted using the Folch method [
27]. According to Costa and Bragagnolo [
24], the fatty acid composition of fish is not altered by the applied microwave energy. However, it is worth noting that MAE can be more suitable than extraction methods employing temperatures above 100 °C, as thermosensitive fatty acids such as EPA and DHA can be affected [
28].
In addition to the high MUFA levels, the nutritional and functional value of the oils obtained from the fish by-products was also highlighted by the contents of DHA (11–13%) and EPA (~10%), two important α-linolenic acid derivatives found in marine fish oils and widely used because of their nutritional, medical, and healthcare value [
29]. According to the Commission Regulation (EU) No 116/2010 on nutritional claims made on foods [
30], the claim “high omega-3 fatty acids” can be attributed to foodstuffs containing at least 80 mg of EPA + DHA per 100 g and per 100 kcal. Therefore, both oils obtained from S1 and S2 complied with this claim since they contained about 6.0 and 4.1 g/100 g of EPA and DHA, respectively. The nutritional quality of these oils was also confirmed by the
n6/
n3 PUFA values, which were below 4 (namely from 0.62 to 0.73), as recommended by the Food and Agriculture Organization (FAO) [
19]. Low atherogenicity (AI) and thrombogenicity (TI) indices and a high hypocholesterolemic index (HI) are considered beneficial for decreasing cardiovascular risk [
31]. As shown in
Table 4, the low atherogenic potential is indicated by AI values below 0.32, due to the greater amount of anti-atherogenic MUFAs and PUFAs compared to pro-atherogenic fatty acids (C14:0 and C16:0). The low thrombogenic potential (TI ≤ 0.26) resulted from the higher proportion of anti-thrombogenic fatty acids (MUFAs and
n3 and
n6 PUFAs) in relation to the pro-thrombogenic fatty acids identified in the oils. In turn, the HI values were equal to or higher than 3.7 (
Table 4) and resulted from a relationship between hypocholesterolemic and hypercholesterolemic fatty acids. As verified for fatty acids, none of these lipid quality indices varied depending on the extraction method. In general, these indices are similar to those described by de la Fuente et al. [
26] for oils from fish by-products.