3. Results and Discussion
3.1 Single-Factor Experiments of Camellia seed scCO2 extraction
The loading of crushed Camellia seeds and the flow rate of CO2 were set fixed in the experiments. Varied extraction temperature and pressure were investigated and discussed in detail.
3.1.1 Effect of Extraction Temperature
Rising of the temperature could enhance both the saturated vapor pressure and the diffusivity coefficient of the solute, resulting in better solubility behavior in scCO2, while at the same time, higher temperature also reduced the density of scCO2 decreasing the dissolving capacity of scCO2 (Bogdanovic et al., 2015; Pilavtepe et al., 2012). These two factors have opposing effects on the oil extraction yield, as a result, an optimum temperature could be predicted in theory. Temperature effects are shown in Fig. 1 (a). As the temperature increased, the oil yield decreased rapidly. This indicates that the optimal temperature might appear lower than 313.15 K. Dissolving power of scCO2 plays an more important role in the extraction process. Higher yield was obtained with longer extraction time.
3.1.2 Effect of Extraction Pressure
As the pressure increased, the density and polarity of scCO2 also increased which led to stronger dissolving capacity indicating higher extraction yield. Pressure effects are displayed in Fig. 1 (b). The oil yield increased gradually as the pressure increased, and the yield increased less at higher pressure. In a longer-time dynamic extraction process, more oils were extracted and the highest yield 22.60% was achieved at 30 MPa and 8 h. Within the first 4 hours, the oil yield increased fast, and after 4 hours, it enhanced slowly in a upward trend.
3.2 Oil physicochemical characteristics
Lower AV and POV represents better quality of oil. These two indicators of all the oil samples were determined and compared in Table 3. AVs of commercial pressed oils are all less than 1, while for samples extracted by n-hexane and scCO2, the values are much higher, distributing between 3 and 1. The AV value contributes to distinguishing sourness tastes. SCCE samples were further investigated in Fig. 2. AVs declined obviously as extraction time increased, e.g., from 3.48 mg/g (0-2 h) to 0.64 mg/g (4-6 h) at the condition of 25 MPa and 333.15 K. While for POV, sample extracted by n-hexane was the highest up to 0.493 g/100 g, indicating that solvent reflux process is unfavorable for the oil quality. For SCCE samples, POV were ranging from 0.012 ~ 0.029 g/100g, significantly lower than other oil prepared samples, indicating scCO2 extraction could be a better preparation method keeping oil quality. After further optimizing the supercritical extraction process, it seems promising to prepareCamellia oil of which AV and POV meet the first-grade standard (GBT11765-2018, AV ≤ 0.50 mg/g, POV ≤ 0.25 g/100g).
3.3 Taste characteristics
3.3.1 E-tongue profiles ofCamellia oils
Fig. 3 shows the radar chart of taste variations determined by electronic tongue. Pressed and n-hexane extracted oils samples are displayed in Fig. 3 (a), and scCO2 extracted samples are shown in Fig. 3 (b) (varied temperature) and Fig. 3 (c) (varied pressure). There were no obvious differences in bitterness, astringency, aftertaste-bitterness (aftertaste-B), aftertaste-astringency (aftertaste-A) and saltiness among these samples. However, for sourness, umami and richness, there were significant differences among these oil samples, especially in the SCCE samples. Negative correlations were found between umami and sourness (R2 = -0.997, p < 0.001), richness and sourness (R2 = -0.912, p < 0.001). E-tongue value of reference tartaric acid solution was set as zero, so oil sourness taste value were all negative. The larger the absolute value, the weaker sourness taste. Sourness taste of three major categories oil samples (in descending order) was: scCO2 extracted oils, n-hexane extracted oil and pressed oils. The highest sourness was found in S13 (-0.99), and the lowest was in S2 (-10.32). For the SCCE samples at 25 MPa and different temperatures, shown in Fig. 3 (b), sourness gradually increased in ascending order: Samples (313.15 K) < Samples (323.15 K) < Samples (333.15 K). While for the SCCE samples at 313.15 K and different pressures (Fig. 3 (c)), sourness was highest in oils at 30 MPa. This may be due to that sourness substances are more easier to be extracted under higher temperature and pressure.
Differences of bitterness and astringency taste responses are relative small. For bitterness, the maximum value was in solvent extracted S6 (-1.05). Pressed oils have the smallest bitterness and astringency among all the samples. Similarly, high temperature (S13-S15) and high pressure (S19-S21) contribute to the astringency taste value.
3.3.2 Discrimination of Camellia oil
Principal component analysis (PCA) is mainly used to exhibit a clear visualization of multidimensional taste profiles in a reduced-dimension plot (Zhu et al., 2020). The relative variances of components suggest the relevant importance of the component expressed as percentages in Table 6. In the results from PCA with eigenvalue greater than 1 (Fig. S1 in supplementary material), the first two principal components (PC1 and PC2) accounted for 65.4% and 22.8% of total variance, respectively, with the total cumulative variance contribution for 88.2%. Although the pressed oils were successfully separated from extracted oils, the difference between SCCE samples and n-hexane extracted samples was not obvious. Fig. 4 illustrates three-dimensional score plot (a) and loading plot (b) in PCA for the difference Camellia oils by the first three principal components (PC1, PC2 and PC3), with the total cumulative variance contribution for 97.4%. The samples are well clustered and no overlap was observed among the three groups, implying the taste of three groups were different. Pressed oils showed relative narrow spatial distribution, while SCCE samples, exhibited a wide range of positive to negative scores, almost splitting into 3 subgroups along both PC1 and PC2 directions: G1 (S13~S15 and S19~S21), G2 (S10~S12), and G3 (S7~S9 and S16~S18). (Fig. S2 in supplementary material ). These three groups well illustrated the scCO2 extraction condition differences among the samples, as G1 represents better solubility of solute in scCO2 due to high temperature (S13~S15), or stronger solvent power due to high pressure (S19~S21); G2 represents medium temperature and pressure (S10~S11); G3 represents relative low temperature and pressure (S7~S9 and S16~S18).
8 kinds of taste loading vectors were shown in Fig. 4 (b). Similar loading vectors suggest redundancy in the potential responses and high loading parameters contribute to discriminating among the samples (Saidi et al., 2018). The umami, richness, saltness, sourness, and astringency were mainly responsible for the oil discrimination in the direction of PC1 , whereas aftertaste-A and Aftertaste-B contributed a lot to the separation along with the direction of PC2, and for PC3, bitterness and astringency play an important role in the classification (Fig. S3 in supplementary material). The pressed, n-hexane extracted and scCO2 extracted samples were mainly differentiated by umami, richness, saltness, aftertaste-A and Aftertaste-B with positive score values along PC1 and PC2. Combined the varied extraction processed, PC2 in general may be responsible for special flavor substances, as pressed and strong scCO2 extraction method offered more abundant flavor chemicals in the oils. These clusters are clearly distinguished from each other (Fig. S2 (a) in supplementary material). To be noteworthy, most PC3 may represent free fatty acid, for the acid values of S6 and S13 were relative large in accordance with their high positive score along PC3. Also, peroxide value of the oil samples may involve in PC1, as their increasing trend is consistent. Not all the data fit these regulations, and the reason is hard to suggest because of the complicated multivariate nature of the score plots.
Thus for the particular dataset, reliable discrimination of oil samples of different extraction method can be illustrated by the e-tongue technique.
3.3.3 Cluster analysis
Hierarchical cluster analysis (HCA) discovered and identified relationships between oil varieties by the distance of taste response of the samples (Zhu et al., 2020). HCA provides insight into the taste profiles by dividing similar samples into groups (clusters) (Liu et al., 2020). In this study, HCA was conducted by Ward’s method for aggregation and the squared Euclidean distance as diversity test. Dendrogram of clustered oil samples base on similarities is shown in Fig. 5. The clustering result is almost the same with that of PCA analysis. All the samples were first divided into two main clusters. Cluster I is SCCE samples and Cluster II involved pressed oils and n-hexane extracted oils. ScCO2 conditions related with solvent extraction power contributes a lot to Cluster I. Cluster II was further divided into two sub-clusters, one was consist of pressed oils and the other was n-hexane extracted oil S6, when the number of overall cluster is set to 5 or more, indicating the difference between the pressed and n-hexane extracted samples is less than the SCCE samples of varied conditions. Thus the HCA method was able to distinguish oil samples of different extraction processes base on the e-tongue profiles.
3.4 Correlation between E-tongue analysis and chemical properties
Acid value (AV) and peroxide value (POV) are key parameters to assessCamellia oil quality and gradation. The possibility of estimating the AV and POV level based on e-tongue profiles processed with Multi Factor Linear Regression Model (MLRM) was evaluated, applying the AV and POV data of Camellia oils produced from different extraction methods determined by titration according to National Standard. The model results were clearly shown in Table 7. Based on 8 kind of potential signals recorded by 5 sensors, which was further selected using stepwise regression method, MLRM models were established with certain ability predicting AV and PV of Camellia oils (AV: R2 = 0.702, p < 0.001; POV: R2 = 0.632, p < 0.001). Lack of the linearity of the models may be due to the inhomogeneity in 1 kg scale scCO2 extraction process which resulted AV and POV fluctuating. Fig. 6 shows “determined value versus predicted value” plot of the MLRM derived for acid value (a) and peroxide value (b). For assessing these physiochemical characterizations using e-tongue technique coupled with chemometrics, satisfactory results were also reported (Rodrigues et al., 2019; Semenov et al., 2019). Although the precision of the models could be further improved, the method still seems attractive for obtaining two key quality values in one measurement. This indicates that e-tongue system would be a promising technique for quantification of Camellia oil quality parameters.