E. oleracea fruits were collected in Abaetetuba (Pará State, Brazil) in the middle of the harvest period (GPS coordinates 1°46’42” S-48°51’07” W). Once collected, the fruits were taken to the laboratory, washed under tap water, and softened in water at 40 ºC for 1 hour. Then, the softened fruits were pulped by manual friction using sieves. Small pieces of the surrounding pulp of the fruits, with an average size of 0.65 cm, commonly called “decocts”, were obtained. Lastly, the decocts were stored at -20 °C until the EAE experiments were performed.

The dry matter content of the decocts was determined according to the NFTA Method 2.1.4 (AOAC Official Method 935.29 & 945.15) by drying the manually-milled wet decocts in a drying and sterilizing stove (SOLAB SL-100, São Paulo, Brazil) at 105 °C for 3 hours (or until constant weight). The result was expressed as the ratio between the mass of water and the total mass of the wet decocts (%, w.b.).

The lipid content of the decocts was determined according to the AOAC Official Method 945.16, by submitting milled dry decocts to Soxhlet extraction (SOLAB SL 145/6, São Paulo, Brazil) using petroleum ether at 60 °C for 3 hours. The result was expressed as the ratio between the mass of lipid and the total mass of the dry decocts (%, d.b.).

The three commercial EP (LNF Latino Americana, Rio Grande do Sul, Brazil), were selected according to the conclusions of Ferreira et al. (2018):

The methodology for the EAE process of açaí oil was based on and adapted from Nascimento et al. (2008). Each essay was performed using 100 g of thawed wet açaí decocts. They were thermally pretreated at 70 °C for 1 min using a domestic food steamer (Walita / Philips Jamie Oliver RI9132/01, São Paulo, Brazil) with the aim denaturing the endogenous enzymes and partially cleave the lignin. Then, the decocts and 240 mL of distilled water at ambient temperature (wet decocts: water ≈ 1:2.5 (w:w)) were transferred to a 600 mL beaker sealed with aluminum foil. The three EP were inserted sequentially using a micropipette in the proportion and the total enzymatic concentration according to the experimental designs. A set of triplicate essays was performed without the addition of EP, considered as the negative control.

For the incubation step, the beaker was incubated at 50 °C under orbital agitation of 100 rpm in a shaker (SOLAB SL-222, São Paulo, Brazil) during an extraction time fixed according to the experimental designs. After the incubation, the beaker was cooled in a bath with ice cubes until reaching ambient temperature.

For the solid-liquid separation step, the beaker was left for approximately 24 hours at ambient temperature in order to obtain natural separation of the solid cake (i.e. fibers and high density proteins), the aqueous-proteic phase (i.e. water and low density proteins) and the lipid phase (i.e. oil droplets dispersed on the surface, optionally in protein emulsion). Note that this step may also be realized using a centrifuge lab equipment. The liquid phases (i.e. aqueous-proteic and lipid phases) were then transferred to another smaller beaker.

For the liquid-liquid separation step, about 20 mL of n-hexane (Dinâmica, São Paulo, Brazil) was added onto the surface with a micropipette in order to favor, on the one hand, the separation of the lipid phase from the aqueous-proteic phase, and on the other hand, the agglomeration of all oil droplets. This strategy was selected as a reference standardized technique at laboratory scale with the aim of focusing only on the effect of the carbohydrase EP (i.e. avoid the use of peptidases) and to allow for the most precise analytical quantification of the extracted oil. Then, the supernatant organic-lipid phase was separated from the aqueous-proteic phase using a micropipette and transferred to a 100-mL previously weighed beaker.

This beaker was incubated (at 100 °C for 15 hours) in a drying and sterilizing stove (SOLAB SL-100, São Paulo, Brazil), until constant oil weight (i.e. complete evaporation of the n-hexane), then stored in a desiccator until reaching ambient temperature. Eventually, the mass of the extracted oil was determined by the gravimetric method, using an Analytical balance (SHIMADZU UniBloc AUY 220, São Paulo, Brazil).

The efficiency of the extraction method, i.e. the oil recovery yield, hereafter called yield (expressed in percentage), was calculated by the ratio between the mass of the oil extracted from 100g of wet decocts by the EAE process, and the mass of the total lipid content in 100 g of wet decocts obtained by the reference Soxhlet method (considered as the positive control).

The identification of the optimal values for the parameters was made through two experimental designs: a three-component Simplex-Lattice Mixture Design for the proportion of the three EP and a two-factor Central Composite Design for the total enzymatic concentration and the extraction time. The ranges of values were determined according to the literature (using scopus data base) related to the EAE process for vegetable oils.

The effect of the proportion of the three EP on the yield was studied using a three-component augmented Simplex-Lattice Mixture Design. The total enzymatic concentration and the extraction time were fixed at 3% and 4 hours, respectively. Our matrix was carried out with the following 5 coded (and uncoded) levels: 0 (0%), 0.33 (33.3%), 0.5 (50%), 6.6 (66.6%) and 1 (100%). The experimental design presents 15 experimental points, in which 3 are single EP essay treatments (essays 1-3), 9 are two-EP mixture essays (i.e. mid-points, essays 4-12) and 3 are three-EP mixture essays (i.e. central point performed in triplicate, essays 13-15) (see Table 1).

The effect of the total enzymatic concentration and the extraction time on the yield was studied using a two-factor Central Composite Design 2² with α = 1.41. Our matrix was carried out with the following 5 coded levels: -1.41, -1, 0, 1, and 1.41. The total enzymatic concentrations investigated were: 0.5, 1.08, 2.5, 3.9, and 4.5% (w:w), and the extraction times studied were: 2, 3.5, 7, 10.5 and 12 hours (see Table 2). The central point (2.5% and 7 hours) was performed in triplicate.

The experimental designs and statistical analysis of the data were performed using the STATISTICA software version 7.0 (Statsoft Inc., Oklahoma, USA). The significance of the models and the coefficient estimates of the factors were evaluated by Analysis of Variance (ANOVA) with a significance level of 5% (α = 0.05). The optimization was performed using the Response Surface Methodology.

The decocts presented a dry matter content ranging from 45.14 ± 0.33% to 46.50 ± 0.51% w.b. This result is in agreement with those obtained (but not presented) by Ferreira et al. (2018) with an average of 50.98% w.b. at the beginning and 45.5% w.b. at the middle of the harvest. No other comparison was possible with the literature because all the authors work from the pre-processed pulp, mainly aqueous juice.

The decocts used throughout our work showed a high variation in the total lipid content, which ranged from 15.20 ± 0.23% to 24.30 ± 0.61% d.b., with an average value of 18.82 ± 3.6% d.b.. This variation in the total lipid content can be justified by the heterogeneity of the vegetable raw material, which may have presented variations in its composition depending on the harvest and the extraction moment itself (Bichara and Rogez, 2011). Our results are slightly lower than those obtained (but not presented) by Ferreira et al. (2018), with an average of 28.33% d.b. and 22.45% d.b., at the beginning and in the middle of the harvest, respectively. These values are significantly lower than those obtained from açaí juice by Nascimento et al. (2008) with 42.6% d.b., Batista et al. (2016) with 45.4% d.b. and Silva et al. (2019) with 49-57%. It is worth mentioning that there are more fiber contents in the decocts than the juice, contributing to the reduction in the total lipid content of the decocts compared to the juice.

Considering the mathematical expression of the yield used in our paper, the yield of the positive control (i.e. using the AOAC Official Method 945.16) was 100%, while the yield obtained experimentally of the negative control (i.e. using the EAE process without EP, with an extraction time of 4 hours) was 29.51 ± 5.92%. This result is similar to the one obtained by Ferreira et al. (2018) under the same conditions with 34.91 ± 11.81% (see Table 1). Note that the yields of the negative control are similar to those obtained by other authors who studied the EAE process for oil from different vegetable matrices (see Table 3).

The matrix of the experimental design and the results of the three-component augmented Simplex-Lattice Mixture Design for the maximization of the yield as a function of the proportion of the three EP is presented in Table 1. One can easily observe a significant increase in these yields, whatever the proportion used, compared to the negative control.

Our experimental results can be directly compared with the ones obtained by Ferreira et al. (2018), who studied the effect of these three EP individually and in combination using iso-enzymatic proportion at an individual enzymatic concentration of 1%, on the yield of the oil extracted from açaí wet decocts. One can observe that our results, all obtained at an enzymatic concentration of 3%, are slightly lower (see essays 1-3, essays 4-6 and essays 13-15, Table 1). One can observe a variability in our results at the central point (see essays 13-15, Table 1), with an average experimental yield of 52.63 ± 3.66%. Such variability may be explained by the use of biological materials, and the independent repetition of all the essays. Note also that the central point provides the maximum experimental yield.

Model fitting. The fitting of the different order models (i.e. linear, quadratic or cubic) to the experimental results of the yields showed that the cubic was the most appropriate one. In particular, the low global p-value (< 0.0028) indicated that the model was significant, the high R² (> 0.90) was satisfactory to validate the significance of the model, the high R² _adj (> 0.83) showed a good relationship between the experimental data and the fitted model and the high lack-of-fit p-value (> 0.6) indicated that the model seemed to accurately fit the experimental data. The significant coefficients (p-value < 0.05) of the factors of the cubic model are those of the linear effects (p-value < 0.0001), followed by the cubic effect (p-value = 0.0026), and the quadratic effect between Ultrazym AFP-L and Viscozyme L (p-value = 0.0415).

Optimal value of the proportion of the three enzymatic preparations. The optimal value of the proportion of the three EP for predicted maximum yield was determined by Response Surface Methodology. In particular, the contoured response surface using the special cubic model clearly showed that the highest yields were achieved with enzymatic mixtures containing more than 25% of Celluclast 1.5L (see red colored region in Figure 1). According to the prediction profiling, the judicious proportion of the three EP which maximized the yield was 45.9% of Celluclast 1.5 L, 25.3% of Viscozyme L and 28.8% of Ultrazym AFP-L (see red vertical line that indicated the optimal value of each EP, in Figure 2). In these conditions, the predicted maximum yield was 54.07 ± 8.91%. Note that this value is comparable to the maximum yield obtained experimentally (see essays 13-15, Table 1).

Understanding of the enzymatic phenomena. Our results demonstrate that Celluclast 1.5L allowed for obtaining higher yield than using Viscozyme L or Ultrazym AFP-L (see essays 1-3 in Table 1). This can be explained by the fact that endo-enzymes randomly cleave internal bonds of the polymer chain, while exo-enzymes act on the end of the polymer chain. In particular, exo-glucanase (e.g. Celluclast 1.5 L) is essential in the breakdown of crystalline cellulose, and is considered an “input enzyme” for the releasing of oil contained in the vegetable cell (Domínguez et al. 1994). However, due to the overlapping of the different polymers of the vegetable cell wall, the accessibility of Celluclast 1.5 L to its specific substrate is intrinsically low. Therefore, enzymatic mixtures containing other class-type enzymes, in particular endo-enzymes, are more competitive. Our results clearly demonstrate that the enzymatic mixtures of two EP containing Celluclast 1.5L with Viscozyme L or Ultrazym AFP-L show yields higher than those obtained with the use of only one EP (see essays 4-5, 7-10 in Table 1, and red color region in Figure 1). This finding can be explained by the fact that these two endo-enzymes easily and randomly act on different polysaccharides of the vegetable cell wall, and in particular generate ends of cellulose chain for the subsequent action of Celluclast 1.5L (Silvamany and Jahim, 2015). On the other hand, as already pointed out by Ferreira et al. (2018), the essays carried out without Celluclast 1.5 L are considered non-advantageous as showing yields lower than those obtained with the use of only one EP (see essays 6, 11-12 in Table 1, and green color region in Figure 1).

Comparison with the literature. Some papers have already studied the effect of several EP, including cellulase, hemicellulase and/or pectinase, individually and in combination (using iso-enzymatic proportion), on the yield of the oil extracted by the EAE process from different vegetable matrices, using similar values for the enzymatic concentration and extraction time to ours (see e.g. Abdulkarim et al., 2006; Zhang et al., 2007; Tabtabei and Diosady, 2013; Mai et al., 2013, and Ferreira et al., 2018, Table 3). As observed, the results obtained by these authors are of the same order as those obtained in our work. Most of authors pointed out that the cellulase and the pectinase are among the main enzymes which contribute to increasing the yield. Moreover, most of the papers concluded that a synergetic action of the different EP on the vegetable cell wall is necessary to maximize the release of oil.

However, to our knowledge, the literature presents very few works related to the optimization of enzymatic mixtures for the aqueous extraction of oil by employing a Simplex-Lattice Mixture Design. One can cite the papers of Dela Cruz et al. (2007) and Silvamany and Jahim (2015) (see Table 3). More specifically, Dela Cruz et al. (2007) observed that ternary mixtures did not provide significantly higher yields than binary mixtures, and concluded that the maximum yield was achieved using a 1:1 binary mixture of xylanase and cellulose (or amylase). Silvamany and Jahim (2015) showed that maximum experimental yield was acheived using a 1:1:1 ternary mixture of cellulase, hemicelulase and pectinase, an a maximum predicted yield with ratio 0.46:0.34:0.2 (celulase : hemicelulase : pectinase). Such an enzymatic proportion is very similar to that obtained in our work.

Table 2 shows the matrix of the experimental design and the results of the two-factor Central Composite Design for the maximization of the yield as a function of the total enzymatic concentration and the extraction time, using the 0.49:0.25:0.28 ternary mixture of Celluclast 1.5 L, Viscozyme L and Ultra­zym AFP-L, respectively. Once more, one can observe a variability in the results at the central point (i.e. total enzymatic concentration: 2.5%, extraction time: 7 hours) (see essays 9-11, Table 2), with an experimental average yield of 57.55 ± 1.19%. One can observe that the results obtained for the essays realized with a total enzymatic concentration of 2.5% and extraction times of 2 and 7 hours (see essays 7 and 9-11, Table 2) are coherent with the one predicted by the previous model for a total enzymatic concentration of 3% and an extraction time of 4 hours. Note that the highest experimental result was obtained for essay 8, with 60.19%, performed using a total enzymatic concentration of 2.5% and an extraction time of 12 hours.

Model fitting. The fitting of the different order models (i.e. including linear, quadratic or interaction effects, and their combination) to the experimental results of the yields showed that the simpler second order models (i.e. that combine the linear & quadratic effects) were the most appropriate ones. In particular, they are characterized by the lowest global p-value (< 0.08), the highest R² (> 0.75), the highest R² _adj (= 0.59), and the highest lack-of-fit p-value > 0.08. This is considered a good result in view of the great variability in handling biological raw materials. The most significant coefficients (p-value < 0.05) of the factor of the second order model are those of the quadratic effect (p-value = 0.015) of the total enzymatic concentration and the linear effect (p-value = 0.019) of the extraction time, followed by the linear effect (p-value = 0.041) of the total enzymatic concentration.

Optimization of the total enzymatic concentration and extraction time. The optimal values for the two factors for predicted maximum yield are identified by Response Surface Methodology. One can easily observe the quadratic effect of the total enzymatic concentration and the linear effect of the extraction time on the yield (see contoured response surface in Figure 3). According to the prediction profiling, the values for the total enzymatic concentration and the extraction time which maximized the yield are 2.85% and 10.9 hours, respectively (see red vertical line that indicates optimum value for each factor, in Figure 4). In these conditions, the predicted yield was 60.55 ± 5.98%. Note that this value is similar to the maximum yield obtained experimentally (see essay 8, Table 2).

As the optimized extraction time lies at the upper extremity of the range of values, some more independent essays were performed in triplicate using extraction times of 8, 12 and 16 hours in order to statistically compare the experimental yields obtained. The results (57.55 ± 1.19%, 57.28 ± 2.77% and 56.03 ± 1.43%, respectively) were analyzed by Analysis of Variance (ANOVA) with a significance level of 5% (α = 0.05) and show that there is no statistical difference between the results (p-values > 0.05).

Understanding of the enzymatic phenomena. As pointed out previously, the yield is significantly influenced by the quadratic effect of the total enzymatic concentration. The yield was enhanced when the total enzymatic concentration increased, because the enzymatic mixture of the three EP hydrolyzed the different polymers of the vegetable cell wall, which enable the release of the oil. However, once the total enzymatic concentration reached the maximum value, the yield slightly decreased with further increase in the total enzymatic concentration. This may be explained by possible enzyme aggregation caused by excess enzymes, which restricted their individual flexibility and activity, by the accumulation of intermediate products which inhibit the enzymatic activity, or by the stabilization of a fat emulsion in water by the enzymes which act as protein surfactants. Regarding the extraction time, the statistical treatment pointed out the significant linear effect of the extraction time, but the prediction profiling, the desirability function and the extra experimental essays showed that an increase in the extraction time up to 12 hours did not provide a significantly (α = 0.05) higher yield compared to 8 hours. Some authors explained that a long extraction time might contribute to the depletion of the substrates and/or the product inhibition of enzymes (Zhang et al., 2007; Li et al., 2011), to a partial inactivation of the enzyme and/or to the adsorption of the oil onto the remaining solid fraction, or to the liberation of cellular components including phospholipids, glycolipids, and proteins, which form lipid-protein aggregates and emulsion stabilization (Nguyen et al., 2020), consequently reducing the yield.

Comparison with the literature. Some papers have already used experimental designs for the identification of the optimal values of the total enzymatic concentration and the extraction time (among others) for the maximization of the yield of the oil extracted by the EAE process from different vegetable matrices, using cellulase, hemicellulase and/or pectinase (see Table 4).

Regarding the total enzymatic concentration, Xie et al. (2011), Mai et al. (2013), Nguyen et al. (2020) and Liu et al. (2020) also observed the significant quadratic effect of the total enzymatic concentration on yield. Other papers (Abdulkarim et al. 2006; Zhang et al. 2007 and Womeni et al., 2008) showed a stabilization of the yield above a value of total enzymatic concentration (see Total enzymatic concentration (%), Table 4).

Regarding the enzymatic time, Zhang et al. 2007, Womeni et al. 2008, Xie et al. 2011, Mai et al. 2013 and Nguyen et al. 2020 also pointed out the significant linear effect of the extraction time, even if they showed that an increase in the extraction time up to the predicted optimal value did not provide a sig­nificant higher yield. This was also the conclusion of the papers of Abdulkarim et al. (2006); Zhang et al. (2007) and Liu et al. (2020) (see Extraction time (hours), Table 4).