RBO (Saman S.A., Uruguay), previously characterized by Benitez et al. [9], was used as a lipid phase. PPC (SolaeTM, TruProTM 2000, Dupont Nutrition and Health) had a protein, fat, ash, moisture and carbohydrate content, according to the label, of 80.6% (w/w), 9.0% (w/w), 4.9% (w/w), 4.5% (w/w), and less than 1% (w/w) respectively. AG (TICAmulsion A-2010, TIC GUMS, Argentina) had a protein, fat, ash, moisture and carbohydrate content, according to the label, of 0.8% (w/w), 2.8% (w/w), 7.7% (w/w), 7.3% (w/w), and 81.2% (w/w), respectively. Finally, MD (Farmal DE-15, Ingredion, Argentina) had a carbohydrate, moisture and ash content according to the label of 94.0% (w/w), 6.0% (w/w) and 0.05% (w/w), respectively. PPC, AG and MD were used as emulsifying and/or stabilizing agents. All the other reagents used were analytical grade.

O/W emulsions were composed of 75.0% (w/w) aqueous dispersions and 25.0% (w/w) RBO [oil mass fraction (Φ_m) = 0.25]. These aqueous dispersions were composed of different mixtures of PPC, AG, and MD to reach a fixed 6.0% (w/w).

The three independent variables in the mixture design were the concentration of PPC (X₁), MD (X₂), and AG (X₃). The resulting augmented {3,3} simplex lattice design calculated with the Minitab Statistical Software (Minitab version 17, USA) resulted in thirteen runs and two replications of the central point (Figure 1).

All tested formulations are shown in Table 1. Before emulsification, PPC, MD, and AG were hydrated with distilled water keeping the aqueous dispersions in magnetic stirring at 1,180 rpm (Wisd, MSH-20D, Germany) for 30 min at room temperature. Then, the amount of RBO needed for each formulation was added. Coarse emulsions were prepared with a high-speed blender UltraTurrax T18 using an S18N-19G dispersing tool (Janke & Kunkel GmbH, Germany) at 18,000 rpm for 90 s. To obtain fine emulsions, coarse emulsions were homogenized with a two-valve high-pressure homogenizer (Panda 2000, GEA Niro Soavi, Italy) at 15 MPa and with three cycles of homogenization. The obtained fine emulsions were stored at refrigeration temperature (4°C) and protected from light. Antimicrobial agents were not added in this study. Three individually prepared replicates were assayed for each condition.

Results corresponding to the particle size distribution, rheological behavior and global stability were fitted using Scheffe’s canonical special cubic model for mixtures of three components (Equation 1), using the multiple-step regression analysis described by Cornell [28]. Equation 1:

Where: Y is the predicted response (in this study: D_[4,3] – D_[3,2], BS_{0 h}, BS_{24 h bottom}, BS_{24 h top}, BS_{72 h bottom}, BS_{72 h top} and apparent viscosity at 200 s^–1), X_i is the proportion of each component (X₁ = PPC, X₂ = MD, X₃ = AG), and β_i, β_ij, and β_ijk were the regression coefficients obtained for each variable corresponding to linear terms, double and triple interaction terms, respectively.

The formulation of the aqueous phase was optimized with Derringer’s desirability function [29] seeking to maximize the response variables associated with the stability of the emulsions (BS_{0 h}, BS_{24 h bottom}, and BS_{24 h top}, BS_{72 h bottom}, and BS_{72 h top}) and to minimize their polydispersity (D_[4,3] – D_[3,2]) simultaneously.

For model validation, O/W emulsions were prepared as was previously described (section Characterization of the O/W emulsions) using the ingredient concentrations (PPC, MD, and AG) that maximized the desirability function, characterized as described in section Surface response analysis and the experimental data was compared to the values predicted by the model.

Results were expressed as mean ± standard deviation. Minitab Statistical Software (Minitab Inc., version 17, USA) was used to perform Analysis of Variance (ANOVA), fit the polynomial equations to experimental data, and obtain the coefficients of such equations. The significance of each term of the models was evaluated and referred to the pure error. For verification of the model adequacy, the lack of fit and the coefficient of determination (R²) were calculated. Minitab was used as well for the numerical optimization procedure through Derringer’s desirability function. The statistical differences among samples were determined using Tukey’s honestly significant difference (HSD) test with a level of signification P = 0.05.

The formulation and the procedure for obtaining the emulsions play a fundamental role in their properties and characteristics. An adequate relationship between the lipid and aqueous phases, the presence of emulsifiers and stabilizers, promotes the stabilization of the emulsions. Additionally, a correct input of processing energy favors a greater increase in their stability over time, due to the reduction and homogenization of particle size distributions [9].

Although there are different methods for studying the emulsions’ stability such as transmission and scanning electron microscopy and fluorescence microscopy [30], in this particular work, the particle size distribution (Figure 2), the mean diameters (D_[4,3] and D_[3,2]) and the difference between them, the flow behavior and the initial stability were characterized immediately after the preparation of the emulsions. In addition, stability measures were made during their sealed refrigerated storage (Table 2). All the emulsions, formulated with the pure components, binary and ternary mixtures, processed after three homogenization cycles, presented multimodal particle size distributions (Figures 2A–C). The particle size range of all processed emulsions was between 0.08–1,000 µm. The emulsions had mean diameter values in the range of D_[4,3]: 1.5 µm to 246.6 µm and D_[3,2] from 0.3 µm to 39.5 µm (Table 2). The smaller differences between the average parameters of particle diameter distribution (D_[4,3] – D_[3,2]) indicate that the system presents a certain homogeneity in size and uniformity in the shape of the particles [26, 31].

The differences (D_[4,3] – D_[3,2]) range from 0.6 µm to 179.0 µm. For all processed emulsions, apparent viscosities were obtained at a shear rate of 200 s^–1, in the range from 11.8 mPa.s to 325.8 mPa.s (Table 2). The 200 s^–1 shear rate is characteristic of some food engineering unit operations such as pumping and mixing [27]. The emulsions studied presented high values of initial backscattering (BS_{0 h} around 80%) throughout the measurement tube (Table 2). This suggests a large concentration of small droplets that are homogeneously distributed throughout the measurement tube [32], except for the formulation with MD (F7) in its aqueous phase composition, which presented a BS_{0 h} around 32 %. After 24 h of refrigerated watertight storage, decreases in the backscattering values of the emulsions were observed, more noticeable in some formulations (Table 2). These decreases turn out to be more evident at the top of the tube than at the bottom. This shows that most systems suffered destabilization mainly due to flocculation and coalescence. After 72 h of refrigerated watertight storage, a slight drop in backscattering values was observed regardless of the formulation and the zone of the tube. However, the emulsions that presented the highest values of backscattering were those that presented mainly PPC or AG in their formulation, allowing the interpretation of their usefulness in emulsifying and stabilizing droplets of the lipid phase.

The formulation of the aqueous phase largely determines the properties of the resulting emulsions, as can be noticed in the formulations prepared. The results presented in Table 2 were statistically analyzed through a mixture experimental design and mathematical models. The coefficients for each response variable are presented in Table 1. All the mathematical models obtained successfully predicted the response variables. The models presented R² greater than 0.74 and a statistically significant P-value (P < 0.05) for the lack-of-fit test, except for differences in D_[4,3] and D_[3,2], BS_{24 h bottom} and BS_{24 h top} (P > 0.05). However, the adjusted R² value is considered a more important parameter to evaluate the models to estimate the response variables.

Each of the regression coefficients (β_i, β_ij, and β_ijk) shown in Table 1 describes the magnitude and direction of the relationship between the term included in the model and the response variable. Also, the model does not include a constant, because it is incorporated into the linear terms. For these reasons, in the linear terms of the components, there are no P-values for statistical comparison [33].

The interaction terms β_i, β_ij, and β_ijk indicate that the association between the mixture of components and the response is statistically significant when P < 0.05. Positive coefficients of the interaction terms indicate that the components of the term act synergistically; while the negative coefficients are related to an antagonistic interaction when the maximization of the response variable is desired [33]. Contrarily, when the response variable needs to be minimized, the interpretation of the coefficient signs is the opposite.

The emulsions formulated only with pure components presented dissimilar values of differences between the parameters D_[4,3] and D_[3,2]. The F7 (MD) presented a value close to 178 mm, while the F1 (PPC) of 30 mm and the F10 (AG) of 1.4 mm. The lower values would indicate a greater homogeneity in the sizes and shapes of the drops. The binary mixtures formulated with AG/DM (F8 and F9) presented the lowest values of the difference between the parameters.

In the binary mixtures of PPC/MD (F2 and F4) intermediate values were observed, while the mixture of PPC/AG (F3 and F6) registers the greatest values of the difference between the parameters D_[4,3] and D_[3,2]. This suggests that the binary mixtures with AG/MD and PPC/MD present a synergy to act at the interface between the aqueous and lipid phases. This behavior is shown in Table 1, where negative values are observed for the PPC/MD and AG/MD coefficients and smaller, positive values for the PPC/AG interaction. The PPC/AG binary mixtures show marked competitiveness at the lipid-aqueous interface. Possibly, the interaction between PPC and AG promotes less homogeneity of sizes and shapes (Table 1).

The ternary mixtures (F5 = F14 = F15, F11, F12, and F13) presented different values between the parameters D_[4,3] and D_[3,2] between 16 mm and 59 mm (Table 2). The value of the coefficient in Table 1 for the PPC/MD/AG interaction presents a positive value, allowing the interpretation that this interaction promotes the heterogeneity of droplet sizes and shapes. However, that ternary formulation where AG predominates in the composition (F13), presented the lowest values in this response variable. This gives rise to the interpretation that in a ternary mixture, the greatest homogeneity of sizes is found with AG as the main component and highlights the good emulsifying properties of this ingredient.

The contour plots of D_[4,3] – D_[3,2] for the O/W emulsions is presented in Figure 3. It can be seen that the greatest differences between the parameters D_[4,3] – D_[3,2] are presented with MD as the main component and in formulations with equal proportions of PPC and AG. The lowest values in this response variable are found with double mixtures between AG and MD, where AG is the majority and the contribution of PPC is minimal.

Regarding the flow properties of the emulsions, Table 2 shows the apparent viscosities (evaluated at 200 s^–1), and Table 1 shows the regression and correlation coefficients of the model and parameters of the statistical analysis. The highest viscosity values were presented by the emulsions formulated with PPC as the main component in a ternary mixture (F11), followed by the binary mixture containing PPC/AG (F3). Interactions probably occur between the PPC and AG macromolecules that increase the resistance to the flow of the emulsions. In general, gums are negatively charged in a wide range of pH while proteins can be positively or negatively charged, depending on the pH of the medium [34]. PPC and AG carried a net negative charge at neutral pH, thus, tend to fully expand due to electrostatic repulsion, occupying maximum volume in solution and offering maximum resistance to flow. A similar effect was observed by İbanoğlu [35] between the whey protein concentrate (WPC) and AG. Therefore, a trend can be observed where the emulsions that present PPC as an ingredient of the aqueous phase show the highest viscosity values, and those others with a lower proportion or absence of PPC as an ingredient, present the lowest values of apparent viscosity evaluated at 200 s^–1.

The coefficients of the PPC/AG interactions are positive (Table 1), allowing the interpretation that the interaction of these two compounds promotes the increase in viscosity. The PPC/MD and AG/MD interactions are negative, acting synergistically to decrease the apparent viscosity. Likewise, the term corresponding to the triple interaction was negative. However, the regression obtained values close to R² (0.41) for model fit and P-values greater than 0.05. Therefore, the apparent viscosity response variable will not be considered for optimization.

The contour plot of the apparent viscosity evaluated at 200 s^–1 is presented in Figure 4. As, it can be seen that the highest values in this response variable are found with double mixtures of PPC and AG, with lower contributions of MD. On the other hand, the emulsions formulated mainly with AG and MD and lower proportions of PPC present the lowest values of apparent viscosity.

In a global analysis of Table 1 regarding the backscattering values, it can be seen that all the coefficients, both the linear terms and the double mixtures, are positive; except for the double PPC/MD mixtures for the time 3 days (BS_{72 h top} and BS_{72 h bottom}). The positive incidence of linear backscattering coefficients is relevant in all cases. However, the incidence of the coefficients is statistically significant only for the PPC/AG interactions, while the PPC/MD and AG/MD interactions are not. The triple interaction coefficients are negative, allowing the interpretation that the mixture of the three ingredients in the aqueous phase is ineffective in stabilizing the oil droplets. Despite this, the triple interaction coefficients are statistically significant only for the response variables BS_{0 h}, BS_{72 h top}, and BS_{72 h bottom}.

In the contour graphs (Figure 5A–E) it can be seen that in those emulsions with a higher proportion of MD as the main ingredient of the aqueous phase, the resulting stability is minimal, regardless of time. However, when the MD is found in a lower proportion, accompanying AG and especially PPC, the emulsions with greater stability are found. In addition, it is observed that those double mixtures between PPC and AG, present reduced stability.

The destabilization produced by the combination of PPC and AG is possibly due to their electrostatic repulsion and their competition for the interphase, and may cause a lesser concentration of emulsifier on the interphase. Coacervation between PPC and AG has been reported [36], although at different pH values. The stability seems to be slightly improved by the addition of MD to the PPC/AG mixture, however, the combination of the three ingredients continues to be worse than the binary mixtures PPC/MD and AG/MD.

To identify a mixture of the aqueous phase that allows obtaining the best properties for O/W emulsions, the optimization was carried out following the desirability function. The desirability function seeks to minimize the values of the parameters D_[4,3] and the difference between the parameters D_[4,3] and D_[3,2]. Additionally, it seeks to maximize the stability values (BS_{0 h}, BS_{24 h bottom}, BS_{24 h top}, BS_{72 h bottom}, and BS_{72 h top}). The apparent viscosity values at 200 s^–1 were not considered as a response variable in the optimization because they presented R² values below 0.70. It should also be noted that no weights were applied in the statistical software to overestimate or underestimate response variables over others. In the global desirability function, for the responses D_[4,3] and the difference between the parameters D_[4,3] and D_[3,2], the objective was to minimize the values, the target was set in 1 and the limit 10. As for the stability values (BS_{0 h}, BS_{24 h bottom}, BS_{24 h top}, BS_{72 h bottom}, and BS_{72 h top}), the objective was to maximize the values, the target was set at 95 and the limit 80.

The statistical software, through the desirability function, calculated the optimal formulations in the aqueous phase, optimizing the properties of the O/W emulsions processed in three homogenization cycles. The optimal formulation of the aqueous phase for the emulsions obtained turned out to be the one with 78% PPC and 22% MD (without AG); achieving a desirability close to 79%. This is in following with every regression coefficient for the ternary mixtures showing significant antagonism between the three components in all the responses studied.

To validate the predicted optimal O/W emulsion formulation (78% PPC, 22% MD, and 0% AG), the emulsion was prepared and tested. The responses predicted by the model and those obtained experimentally for the optimal O/W emulsion are presented in Table 3.

This emulsion presented a multimodal particle size distribution, with a major peak around 0.8–0.9 µm, the following peak around 34 µm, and a smaller one at 0.2 µm. The values of the parameters D_[4,3] were in the order of 10.66 µm and the difference between the parameters D_[4,3] and D_[3,2] was close to 9.9 µm. These values turned out to be higher than those predicted by the model (Table 3). The optimal emulsion presented experimental values of backscattering regardless of the zone and time that ranged between 87–88%. However, the values predicted by the model (96–98%) are higher than the experimental ones. Even though the actual values of backscattering differ, their variation over time is not significant either in the model or in the experimental verification (Table 3). Therefore, despite the numerical differences, the model obtained allows us to describe, with some limitations, the behavior of the O/W emulsions studied.