Three commercial extracts of Brazilian propolis (brown propolis, red propolis and green propolis) were kindly provided by Beeva Indústria Comércio e Exportação de Mel e Derivados S/A (Marechal Deodoro, AL, Brazil). The green and brown propolis used in the manufacture of the extracts were collected in February 2020 and April 2020, respectively, in Alterosa, MG, Brazil (21˚15'56.2"S 46˚08'17.7"W for green propolis; 20˚48'30.0"S 45˚18'42.3"W for brown propolis). Red propolis was collected in April 2020 in Porto de Pedras, AL, Brazil (9˚09'33.6"S 35˚18'18.7"W), and Canavieiras, BA, Brazil (15˚39'55.6"S 38˚57'01.6"W). According to Beeva S/A, these extracts were produced in their commercial manufacturing plant by maceration using hydroalcoholic solution (70% food grade ethanol) as the extracting solution for 7 days. In Brazil, a comprehensive list of solvents used in propolis extract production is regulated, including acetone, ethanol, isopropanol, hexane, methanol, and hydroethanolic solutions [33] [34] , and solvent selection depends on prior analysis of multiple factors, such as propolis composition, yield, price, and technological trends [35] [36] [37] . Hydroethanolic solutions of ethanol:water of 70:30 (v/v) are conventionally used at a commercial scale to manufacture propolis extracts [35] [36] [38] due to the low toxicity, obtaining a wide range of bioactive compounds [36] . Ethanol is a readily available, low-cost solvent that is generally recognized as safe (GRAS) [37] . Other green solvents, such as water, limonene, and vegetable oils, have been seen as alternatives to replace ethanol-based extracts and other organic solvents, even though they have lower extraction performance [39] [40] [41] .

The extracts were characterized (Table 1) for the content of total phenolic compounds and total flavonoids, by high-performance liquid chromatography-diode array detection (HPLC-DAD), and the separation was performed on a reverse-phase C-18 hypersil gold column (5-μm particle size, 150 mm, 4.6 mm; Thermo Fisher Scientific, Massachusetts, USA) at 38˚C [42] , and dried extract by drying in an oven with air circulation at 102˚C ± 2˚C to constant weight [43] .

A SLMD was used to find out the interactive effects of antioxidant activity from the red, green and brown PE in the blend extracts [44] [45] . This system has symmetrically distributed experimental points and a well-chosen polynomial equation to represent the surface response over the entire SLMD region. The distribution of the independent variables (x₁, x₂ and x₃) in the simplex space (equilateral triangle), Figure 1, the sum of all must be 100% ( ∑ X n = 1 ). The distribution of mixtures within this space is known as lattice, which a mixture can be pure element (triangle vertex) or binary interactions (triangle edges) and ternary (triangle center) between them. In this trial, the SLMD for 3 factors and a 3^rd degree polynomial will consist of ⁵C₄ = 9 points plus one central point. Therefore, the design points are: (1, 0, 0), (2/3, 1/3, 0), (1/3, 2/3, 0), (0, 1, 0), (0, 2/3, 1/3), (0, 1/3, 2/3), (0, 0, 1), (1/3, 0, 2/3), (2/3, 0, 1/3) and the central point with equal parts of all components of the mixture (1/3, 1/3, 1/3). Therefore, 3 points of pure PE (vertex), 6 points of blend PE (edges) and one central point (inside of the triangle) with 4 replications are obtained (Figure 1). The design of experiment (DOE) is shown in Table 2. The cubic polynomial RSM in mixture design model is presented in Equation (1).

Y i j k = ∑ i = 1 3 β i X i + ∑ i ≤ j 3 ∑ j 3 β i j X i X j + ∑ i ≤ j 3 ∑ j ≤ k 3 ∑ k 3 β i j k X i X j X k + ε i j k (1)

where the Y i j k is observed value in the i-j-k-^th propolis extract level; X i , X j and X k are coded independent variables; β i , β i j and β i j k are linear, quadratic (component pairs) and cubic (3-component sets) regression coefficients of the model, respectively, whose quadratic and cubic parameters may be either

¹Treatments; ²CP = central point; ³PE = propolis extract.

synergistic or antagonistic blending; and ε i j k is the error associated to the Y i j k observation. Note that the intercept term is not included due to the correlation between all the components (their sum is 100%). The effect of blends PE on the antioxidant activity (trolox equivalent antioxidant capacity, TEAC/mg dry weight) was analyzed using the least square method. For this trial, typical multiple regression models were used: linear model, Equation (2), quadratic model, Equation (3), and special cubic model, Equation (4).

Y i j k = β i X i + β j X j + β k X k + ε i j k (2)

Y i j k = β i X i + β j X j + β k X k + β i j X i X j + β i k X i X k + β j k X j X k + ε i j k (3)

Y i j k = β i X i + β j X j + β k X k + β i j X i X j + β i k X i X k + β j k X j X k     + β i j k X i X j X k + ε i j k (4)

where the β’s is the model parameter coefficient; X’s are the independent variables (input points); Y i j k and ε i j k as previously defined. The selection model was based on ANOVA and significance test and, for the lack of fit and evaluation of the coefficient of determination R². Antioxidant capacity values were explained by the three-dimensional response surface analysis of the independent variables. From the predictive model, a mixture of propolis extracts with optimal antioxidant activity was established. The adjusted validation model was determined comparing the predict antioxidant activity from the optimized mixture with experimental data by student’s t-test. Statistical analyzes were carried out using the Statistica version 9.1 software (StatSoft Inc., Tulsa, OK, USA) at 5% significance level.

The antioxidant capacity of propolis extracts was evaluated through oxygen radical removal capacity (ORAC) fluorimetric method using fluorescein as a molecular probe and 96-well black microplate [46] . The ORAC method is based on the elimination of peroxyl radicals (ROO•) by the antioxidant, by the hydrogen atom transfer mechanism (HAT), under controlled temperature, pH and time (Figure 2). The radical ROO• is produced during the assay by the thermal degradation of a radical generator, usually an azo compound, including α,α-azobisisobutyronitrile (AIBN), 2,2’-azobis(2,4-dimethylvaleronitrile) (AMVN), 2,2’-azobis(2-amidinopropane) (ABAP), and the hydrophilic 2,2’-azobis(2-amidinopropane) hydrochloride (AAPH). Since ROO• is one of the main free radicals produced during the oxidation of biological systems, the data obtained through this method are considered representative for food systems [47] [48] .

The propolis extracts and blends were diluted with ethanol 99.8% (1:50, v/v) and 75 mM phosphate buffer, pH 7.4 (1:400, v/v). The total dilution factor was 1:20.000 (v/v). A blank sample (70%, v/v, hydroalcoholic ethanol solution) was diluted as previously described for the propolis extracts. Based on the reference method, 25 µL of diluted extracts, blends or blank and 150 µL of fluorescein solution (81 nM in 75 mM phosphate buffer, pH 7.4) were pipetted into each well of the microplate. After incubation (10 min at 37˚C) under constant agitation, 25 µL of 2,2’-azobis (2-amidinopropane) dihydrochloride (AAPH) solution (152 mM in 75 mM phosphate buffer, pH 7.4) were added to each well of the microplate. The decay of fluorescence was followed every minute of reaction at 37˚C between 0 and 90 min, under excitation wavelength of 485 nm and emission wavelength of 528 nm. The result was expressed as TEAC/mg dry weight, based

on a calibration curve prepared using trolox solutions with concentrations between 2 to 128 µM, diluting the standard trolox solution in 75 mM phosphate buffer, pH 7.4 [46] .

Regression model obtained through the SLMD was used to find out the interactive effects of antioxidant activity from the red, green and brown PE in the blend extracts. Predictive model signs coefficients are evaluated for identification of favoring (synergism) or compromise (antagonism) in the relationship between two or more predictor variables: coefficients with positive signs indicate synergistic effect between the predictors, while negative signs indicate antagonism. The absence of polynomials of degree 2 or higher in the predictive model indicates an additive relationship [12] [49] . Synergism values (SV, %) were calculated as Equation (5). Theoretical ORAC was calculated as the weighted average of the observed antioxidant activities for each component extract of the mixture. The obtained difference values (%) indicate that there are potential synergistic (positive values) or antagonistic (negative values) and additive (null values) effects [50] .

SV ( % ) = [ experimentalORAC theoreticalORAC × 100 ] − 100 (5)

An adequate model must be significant (p ≤ 0.05) and with non-significant lack of fit (p > 0.05) [45] [51] . The coefficient of determination informs about the model predictive capacity. When the value is closer to 1.00, better is its capability. R² values ≥ 0.70 indicate high correlation and it is considered sufficient to demonstrate the adequacy of the model [52] [53] . The quadratic model was the unique model that met the three criteria (Table 3), and then was selected. However, the model was reparametrized, eliminating the non-significant coefficients (p > 0.05) (Table 3), obtaining the following adjusted equation for antioxidant activity, Equation (6):

y ^ ORAC ( TEAC / mg ) = 4.0500 x 1 + 6.0593 x 2 + 3.9917 x 3 + 12.1394 x 2 x 3 ( R 2 = 0.81 ) (6)

A ternary response surface diagram was obtained (Figure 3). It is observed that the response for antioxidant activity was greater around the edges x₂x₃, represented by the binary mixture of green PE (x₂) and brown PE (x₃) and intensified as the proportion of each extract in blend approached 1:1 (midpoint of the vertex). In fact, the adjusted equation revealed a binary mixture formed by 59% of green PE and 41% of brown PE as maximum point of antioxidant activity, whose predicted value was 8.15 µmol TEAC/mg dry weight.

¹red, ²green and ³brown propolis extracts, x₁ x₂ and x₃ are coefficients of linear terms; x₁x₂, x₁x₃ and x₂x₃ are the coefficients of quadratic terms; and x₁x₂x₃ is the coefficient of special cubic term. ^nsNot significant (p > 0.05); *Significant (p ≤ 0.05); **Significant (p ≤ 0.01); ***Significant (p ≤ 0.001).

The response observed experimentally for the optimized mixture (8.22 ± 0.75 μmol TEAC/mg dry weight; n = 5) did not differ (p > 0.05) from the predicted value (8.15 µmol TEAC/mg dry weight), ensuring the suitability of the adjusted predicted model for the antioxidant activity of the mixtures.

Synergism studies are important to understand how the combination of antioxidants affects the activity of each component and the overall consequence for the mixture in terms of efficiency. It is known that these mixtures of ingredients can result in more or less efficiency, leading for possible synergistic or antagonistic relationships. In synergistic relations, the effect is greater in the mixture than the sum of the isolated effects. Although, when the response in the mixture is smaller than the sum of each isolated effect, the effect is antagonistic. In the absence of these effects, it can be said as additive when the result is the same as the sum of each isolated effect [11] [12] .

The occurrence of term x₂x₃ (+ 12.1394x₂x₃) in regression model indicates that there was a synergistic interaction between the green and brown propolis extracts, improving 54% (SV) the efficiency of the 59% of green PE and 41% of brown PE blend (8.15 μmol TEAC/mg dry weight) when compared with the equivalent sum of isolated effects (5.30 TEAC/mg dry weight). This association is visualized on the vertex x₂x₃ (Figure 3), an increment of antioxidant activity was observed in the center of the edge (blends 1:1 of green PE and brown PE extracts) red region, in relation with the vertex (pure extracts) green/yellow region. The absence of the terms x₁x₂ e x₁x₃ evidences the inexistence of synergy or antagonism between red PE (x₁) with green PE (x₂) or brown PE (x₃), which means that among the other extracts, in any proportions, the relationship was additive. When mixing red PE with green PE or red PE with brown PE, the antioxidant properties are preserved.

Evidence on antioxidant synergism in propolis extract blends is scarce in the literature and available data are based on in vitro studies. To the best of our knowledge, this is the first study that shows a synergistic relationship between Brazilians green PE and brown PE. Mixtures of Portuguese propolis extracts collected over five years (2011-2015) in Gerês, Portugal (41˚45'41.62"N; 7˚58'03.34"W), were evaluated in terms of its antioxidant activity by Peixoto et al. [9] . The authors concluded that the mixtures presented greater antioxidant activity than the individual extracts, and attributed this improvement to the synergism between the different compounds present in extracts, not only limited to phenolic compounds and flavonoids [9] . In addition, Oliveira [54] observed synergistic and antagonistic relationships combining propolis extracts commercialized in different regions of Portugal.

It is believed that the synergistic effects between propolis extracts on antioxidant activity are a function of their complex phytochemical composition, which contribution of minority compounds would be as important as of majority compounds [4] [9] [55] . The structures of phenolic compounds, the main class of bioactive compounds found in propolis, including some markers found in Brazilian propolis [2] [14] [17] [19] [24] [56] , are presented in Figure 4. Flavonoids are one of subclasses of phenolic compounds mostly found in propolis extracts [2] [4] [57] . Other groups of bioactive substances were identified including terpenoids [1] , benzophenones [55] and chalcones [56] .

Souza et al. [57] purified samples of Brazilian brown propolis, obtaining fractions enriched in phenolic compounds and flavonoids. In opposite of expectations, the antioxidant activity of the enriched fractions was lower than the original extract, indicating that there is synergism among different compounds. When investigating the effect of seasonality on the composition and antioxidant activity of Brazilian red propolis, do Nascimento et al. [55] attributed the obtained synergistic effect to interactions between flavonoids and gutiferones and between total phenolic compounds and gutiferones.

Osés et al. [4] compared the antioxidant activity of propolis extracts with analytical standards blend composed of the same majority of phenolic compounds identified in the extracts at equivalent concentrations. The authors reported that the antioxidant activity of the extracts was superior to the analytical standards blend, attributing to the synergistic effect of the minority compounds for better antioxidant activity in the propolis extracts.

Synergism between different classes of compounds present in propolis extracts was observed such as between flavonoids and benzophenones [55] , terpenes and polyphenols [58] and phenolic acids and flavonoids [59] , which may give an idea of the possible interactions that lead to the enhancement of antioxidant activity in the studied propolis extracts blends. The explanation of the compounds and mechanisms involved in these interactions will complement our findings and guide future research for the rational development of new antioxidant mixtures based on these natural resources, with better applicability as natural additives in food, among other commercial applications.