Olive fruits (Olea europaea L.) from the Edremit cultivar grown in the Mut area were harvested in the 2015 crop season. Olives with a maturity index of 3.35 were picked at their early ripening stage in October. The maturity index was calculated according to the method of the International Olive Oil Council (IOOC, 2011). According to this method, 100 fruits were randomly taken to assess their level of maturity by a subjective evaluation of the color of the olive skin and flesh. Then the olives were distributed into eight groups, according to the following characteristics: bright-green skin (group N = 0), greenish-yellowish skin (group N = 1), green skin with reddish spots (group N = 2), reddish-brown skin (group N = 3), black skin with white flesh (group N = 4), black skin with < 50% purple flesh (group N = 5), black skin with > 50% and < 100% purple flesh (group N = 6), and black skin with 100% purple flesh (group N = 7). Maturation indexes ranged from 0 (intense green skin) to 7 (black skin and 100% purple flesh). The maturity index was calculated by Σ(N_in_i)/100, where N is the group number and n is the olive amount in that group (Espínola et al., 2009).

The olives were put in 25 kg boxes and brought to the pilot plant. The total oil content (TOC) and moisture content of the olive samples were determined. Moisture content (% weight/weight) was determined at 120 °C to its constant weight and TOC (% weight / weight) of the olive samples was determined using Soxhlet extraction.

A laboratory mill, equipped with a mixer, a basket centrifuge and a metal crusher was used to extract oil from olive fruits. After the washing process, the olives were crushed and the olive pastes (2 kg each) were sonicated from 2 to 10 minutes and then malaxated for 30 to 50 minutes. The ultrasound system used in this study was ultrasonic bath (150kW and ultrasonic frequency 37±2 KHz; Tank volume 4 L). Each olive paste sample was centrifuged for 1 minute and the oil was collected in 500 ml dark colored bottles. No co-adjuvant or pre-heating process was applied during extraction and all experiments were performed in triplicate.

The temperature measurements were taken by a K type thermocouple by placing the 3 probes at different points of the ultrasound tank for every 30 seconds. The ultrasound tank and the laboratory mill, equipped with a mixer, a basket centrifuge and a crusher were subjected to the cleaning process with water before and after every measurement to avoid contamination which can affect the accuracy of extraction yield and quality characteristic measurements. After cleaning all parts of the extraction unit with water, a thorough drying of the unit was carried out.

Extraction yield was defined as the percentage of the extracted olive oil from the total weight of the fruit (g). The extraction yield was calculated using the formula below:

The ultra-violet absorptions (K232 and K270), the olive oil free acidity (% oleic acid per 100 g of olive oil) and peroxide value (mEq O₂/kg oil) were used for evaluating the qualitative properties of the oils (European Union Commission Regulation, 1991), whereas the total chlorophyll, total carotenoid, total phenol content and total antioxidant activity (DPPH assay) analyses were used to determine the nutritional characteristics of the oils.

The chlorophyll and carotenoid contents were determined colorimetrically at 670 and 470 nm, respectively, and the results were expressed as mg/kg (Gandul-Rojas and Mínguez-Mosquera, 1996).

where A is the absorbance and d is the spectrophotometer cell thickness (1 cm).

Phenols were extracted from the olive oil by liquid–liquid extraction using methanol as solvent and the concentration was measured using the Folin-Ciocalteau reagent at 725 nm. The total phenolic content was expressed as mg of gallic acid equivalents per kg of extra virgin olive oil (Montedoro and Servili, 1992).

The antioxidant activity was determined by the 1,1-diphenyl-2-picryl hydrazyl (DPPH) radical scavenging activity according to the method of (Sevim et al., 2013) with some modifications. 1 g of oil was dissolved in 5 mL methanol and then shaken for 5 min with a homogenizer. The solution centrifuged at 3500 rpm for 10 min in order to separate non-polar and polar compounds. DPPH radical scavenging activity determination was carried out at the methanol phase. The extracts (100 μL) were gently mixed with methanol (1,900 μL) and a 100 mM DPPH solution (1,900 μL) for blanks and allowed to stand in the dark for 15 min at 25 °C. A spectrophotometer (Shimadzu Spectrophotometer UV- 1700 PharmaSpec, Japan) was used to read the absorbance of the resultant solution at 517 nm against a blank sample without DPPH. The Trolox equivalent (TE) of the DPPH radical scavenging activity quantification was determined against a minimum of a five-point calibration curve prepared with known concentrations of Trolox (Sigma-Aldrich Co. LLC, St. Louis, MO, USA). The results were expressed as μmol TE of 100 g of each sample based on a calibration curve (R² = 0.998).

To determine the long term effects of ultrasound, the olive oils were stored at room temperature for 12 months. A sensory analysis was conducted by a panel test according to the method of Achat et al. (2002). The sensory quality of the oil samples were evaluated by 10 trained panellists. The panelists evaluated the overall acceptability of each oil sample (taking into account any positive and negative taste), using a numerical scale 1–5 (1 = not acceptable, 5 = extremely good), as well as bitterness and fruity etc. (1 = no bitterness, 5 = extremely bitter). EVOO classification establishes a mean value of the defects of 0 and a mean value for fruity above 0 according to EU legislation (Achat et al., 2012; Puértolas and Martínez de Marañón, 2015).

A three-level, three-factor, Box-Behnken Design (BBD) was applied to generate a total of 17 experiments for the optimization of ultrasound assisted extraction parameters and to evaluate the combined effects of three independent variables. The ultrasound time (2–10 min.), ultrasound initial temperature (20–50 °C), and malaxation time (30–50 min.) were the parameters and levels employed. The original values and the coded values of independent parameters which were based on the results of our preliminary study are illustrated in Table 1. In order to evaluate the pure error and the curvature the complete design consisted of 17 combinations with 5 replicates of the center point.

The effects of different ultrasound parameters on the qualitative and nutritional parameters of olive oils one way ANOVA were tested on all individual data groups to determine the significant differences using SAS 9.2 (SAS Institute Inc., Cary, NC, 2009). A Tukey’s multiple comparison tests, at p < 0.05, was applied to determine differences among data groups. Correlation coefficients between instrumental results were also determined using SAS.

For the optimization part of this study, Design-Expert version 8.0 (Stat-Ease, Minneapolis, MN) was used to analysis of variance (ANOVA) tables and the regression coefficients of individual linear, quadratic and interaction terms. The experimental data were evaluated to fit a statistical model (Linear, Quadratic, Cubic or 2FI). The coefficients of the model were represented by constant terms, A, B and C (linear coefficients for ultrasound time, ultrasound temperature and malaxation time, respectively), AB, AC and BC (interactive term coefficient), A², B² and C² (quadratic term coefficient). Correlation coefficient (R²), adjusted determination coefficient (Adj-R²) and adequate precision were used to check the model adequacies; the model was adequate when its P value ˂ 0.05, lack of fit P value ˃ 0.05, R² > 0.9 and Adeq Precision > 4. Differences among means were tested for statistical significance using analysis of variance (ANOVA). The statistical significance level was set at P < 0.05.

In order to achieve the lowest acidity and the highest yield, the Box–Behnken design was used to optimize extraction parameters including ultrasound time, ultrasound temperature and malaxation time (Table 2). For the three factors, the Box-Behnken design was chosen since it offers advantages in requiring a fewer number of runs. Results showed that the yield ranged from 3.60 to 8.25% and the maximum extraction yield of olive oil from ultrasound assisted extraction was obtained at higher temperatures and longer malaxation times as reported by several researchers (Angerosa et al., 2001; Inarejos-García et al., 2009; Samaram et al., 2015; Taticchi et al., 2013). These values were relatively lower than the values reported by Clodoveo et al., (2013) and Jimenez et al., (2007). This difference for yield could be related to different varieties of olives and the pre-treatments like heat treatment used in previous studies.

To show the significant (p < 0.05) interaction effects of ultrasound extraction parameters on the yield (Fig. 2a–c) and acidity value (Fig. 2d–f) 3D response surface plots were created. According to the results, a higher amount of oil was obtained by the ultrasound extraction at elevated ultrasound temperatures for longer malaxation time which could confirm that the ultrasound extraction at higher temperatures could disintegrate the cell wall and membranes, and promote the mass transfer from olive fruit (Samaram et al., 2015). However, the ultrasound time applied in the extraction had no significant effect on the yield (Figures2a and 2b). The ultrasound extraction at higher temperatures with longer sonication times resulted in higher acidity (Figure 2d).

The P value, F value, R², Adj-R² and Adeq. Precision estimate of the adequacy of each model is shown in Table 4. High F values (27.90 for yield, 11.16 for AV) and low P values (0.0001 and 0.0022, for yield and AV, respectively), indicated that the models were highly significant for yield and AV. The R² of the predicted models for the yield and AV in olive oil extraction were 0.9729 and 0.9349, respectively. Adj.R² values were 0.9380 and 0.8511 for yield and AV, respectively, which indicated a high degree of linear fit between the predicted and experimental values. The lack-of-fit test showed the fitness of the model. The P value of 0.0705 for the yield and 0.0996 for AV indicated that the models fit the experimental data accurately. These results indicate that the models for AV and the yield were adequate for predicting within the range of the variables employed. The coefficients of the variables in the regression models for the yield and AV and their significance are shown in Table 5.

The free acidity value of olive oil is considered to be a significant factor which determines the quality of oil. The acidity of the oils for all experiments was below the legal limit (< 8 g oleic acid/kg oil) established for the category of EVOO (European Union Commission Regulation, 1991). The low acidity levels found after sonication were presumably associated with a high quality of olive fruits and appropriate storage conditions of the oil until analysis. In terms of AV, the independent variable (B), the quadratic term (B²) and the interactive terms (AB, BC) were all significant (P < 0.05). The quadratic regression model for AV was as follows:

In terms of yield, the independent variables (B, C), the quadratic term (B²) and the interactive terms (AB, BC) were all significant (P < 0.05). The quadratic regression model for yield is given below:

The most significant effect on the extraction yield and the acidity (p ˂ 0.05) was the malaxation temperature among all ultrasound extraction variables. Conversely, ultrasound time showed no effect (p ˃ 0.05) on the yield and the acidity.

In order to optimize the UAE variables which resulted in the most desirable responses numerical optimizations were applied. The purpose of this study was to improve the oil extraction resulting in the highest yield and the lowest acidity. The interaction effects of ultrasound extraction variables on the yield and acidity of olive oil are shown in 3D surface plots (Figure 2a-c and Figure 2d-f). The ultrasound time of 2 minutes, the ultrasound initial temperature of 50 °C and the malaxation time of 43.23 minutes were found to be the predicted optimal conditions of the UAE process of extra virgin olive oil. Under these optimized experimental condition, the predicted values for olive oil yield and acidity were 8.42 % and 0.283 g oleic acid/100 g olive oil, respectively, whose desirability values were equal to 0.914.

To verify the reliability of the models under the optimal conditions, experiments were performed. The experimental values for the acidity 8.31 % and the yield were 0.31 g oleic acid/100 g olive oil. These results were in good agreement with the predicted values under the optimal working condition. Table 6 shows the error rates between the experimental and predicted values which are smaller than 10%. Therefore, the acidity value of olive oil and yield for any combination of ultrasound time, ultrasound temperature and malaxation time could be accurately predicted by the regression models obtained by RSM.