Linear programming formulation of a dairy drink made of cocoa , coffee and orange by-products

Quijano-Aviles, et al.: Agroindustrial by-products Emir. J. Food Agric ● Vol 28 ● Issue 8 ● 2016 555 Linear programming is an important tool that is aimed to maximize or minimized a linear function taking account linear equality and/or inequality constraints. Most of the linear programming models have been used in diet optimization, based on the impact of cost and nutrition as constraints (Moraes et al., 2012). However, some research have demonstrated the importance of considering the palatability of the product as a constraint (Darmon et al., 2006). On this background, we have used the sensorial evaluation to define the constraint of a programing model to optimize a drink formulation in dependence of the polyphenol content of cocoa bean shells, coffee silver skin and orange peel. MATERIALS AND METHODS Extract preparation Fresh samples of cocoa bean shells, coffee silverskin and orange peel were dried individually at 60oC in an oven with air circulation during 16 h. The dried residues were mixed according to the formulations obtained from the experimental design; a 10% aqueous extract was prepared by boiling the mixture at 100oC for 5 min. Higher levels of aqueous extract provided too much astringency to the infusion, resulting in a unpleasant taste. The preparations were allowed to sediment for 5 min at room temperature and the solids were separated by filtration. Experimental design An experimental design was obtained applying the equation 1: Yi = β1X1+ β2X2+ β3X3+ β12X1X2 + β13X1X3+ β23X2X3+ βijXij 2... (1) Here, Yi is the result; the beta symbols (β) represent the coefficients to be adjusted by regression analysis; X1, X12, Xij2 are the lineal, interaction and quadratic effect of the components respectively, in which X1 represent the cocoa bean shells, X2 represent coffee silverskin and X3 represent orange peel. The experiments were performed according to the following restrictions: 74.00% ≤ X1 ≤ 100% (2) 00.00% ≤ X2 ≤ 24.50% (3) 00.00% ≤ X3 ≤ 1.50% (4) The restrictions of the model were defined based on previous experiments where the combination of those components displayed a pleasing flavor. Due to the restrictions applied, the simplex region was sub divided into a new region in which were included the possible mixture combinations as described by Silveira and Leite (2010). In this case, the design of extreme vertices was applied to the experiments with mixtures. Dairy drink preparation Preliminary experimental work was realized in order to define the composition of the drink through trial and error assay. The final acceptable formula was mixing the aqueous extract from by products to skim milk (37%), sugar (5.87%), stabilizers and preservatives (0.13%) and pasteurized at 71.7oC for 15 seconds, followed by fast cooling at 4oC for storage and later evaluation. Sensorial evaluation The sensorial panel was composed of 30 semi-trained participants who were asked to evaluate taste, odor, color and appearance of the samples. Sensorial evaluations were done applying the test of multiple comparisons (ABNT, 1995). Samples were coded with random numbers of 4 digits and compared to a standard labeled as ‘R’, a cold commercial drink of mocha coffee with characteristic similar to the drink we aimed to design. Panelists compared each coded sample with the reference sample and graded them using a sensorial scale of 9 points (Table 1). The data was analyzed applying a variance analysis and Dunnett method for multiple comparisons (Dunnett, 1955). Total polyphenol content and DPPH assay Total polyphenol content (TPC) was determined by the Folin-Ciocalteu method and the polyphenol activity was expressed as Gallic Acid Equivalent (GAE) (Lachman et al., 1998). Briefly, 250 μl sample were mixed with 250 μl Folin-Ciocalteu’s reagent (Sigma-Aldrich). Na2CO3 20% (750 μl) was added after 5 min and the mixture allowed to proceed for 2 h; absorbance was measured at 765 nm. The antioxidant activity (AA %) of each component was determined by the free DPPH assay with some modifications. The reaction mixture consisted of 0.5 ml sample, 3 ml absolute ethanol and 0.3 ml of a 0.5 mM DPPH in ethanol. Absorbance was measured at 517 nm Table 1: Sensorial evaluation and scale of rating scores (ABNT, 1995) Rating score Equivalent rate 1 Extremely inferior to R 2 Markedly inferior to R 3 Moderately inferior to R 4 Slightly inferior to R 5 Equal to R 6 Slightly superior to R 7 Moderately superior to R 8 Markedly superior to R 9 Extremely superior to R Quijano-Aviles, et al.: Agroindustrial by-products 556 Emir. J. Food Agric ● Vol 28 ● Issue 8 ● 2016 after 30 min incubation. Concentration was determined applying the following equation: % inhibition = control – test/control × 100 (5) Caffeine analysis by HPLC The assay for caffeine determination was adapted from the method described by Amaro et al., 2008. A Perkin Elmer 200 instrument was used for high pressure liquid chromatography (HPLC), equipped with UV-VIS detector and a C18 column 250 x 4.20 mm with a solid phase of 5 μm. The mobile phase used was a 40:60 mix HPLCgrade methanol-water, with a flow rate of 1ml/min, 5 min ejection time and detection at 254 nm. Injection volume for samples and standards was 10 μl. Optimization A lineal programming model was developed with the aim to maximize the total polyphenol content in the formulations, limiting this approach to those formulations that scored similar to the reference in the sensorial evaluation. The target objective function was:


INTRODUCTION
Industrial production, including the agricultural industry, generates significant amounts of residues and by-products (Chanakya and Alwis, 2004).Cocoa bean industry dispose tons of cocoa bean shell as waste every year (Arlorio et al., 2005).However, this residue can provide a ready source of inexpensive polyphenols (Redgwell et al., 2003).The presence of flavonoids in extract of cocoa bean shell have been identified (Azizah et al., 1999) and in vitro research have demonstrated that flavonoids prevent low-density lipoprotein oxidation (Morel et al., 1996).Coffee silverskin is one of the main residues of coffee production and can cause serious environmental problems to soil (Nabais et al., 2008).Nevertheless, current investigations has proposed coffee silverskin as a good source of nutrient due to its antioxidant activity (Borrelli et al., 2004).Orange juice is one of the most consumed beverages today (Martin et al., 2010), the generation of orange waste is estimated between 15 to 25 million tons per year (Marín et al., 2007) in which citrus peel is the major constituent, representing 44% of the weight fruit mass (Widmer et al., 2010).The extraction of phenolic compounds from orange peel to use them as antioxidant in foods have been demonstrated (Huang et al., 2009).
Recent research has proposed the use of residues and by products for the formulation of food products (Ribeiro et al., 2014).Main attention has been focused in the use of coffee skin and Martinez-Saez et al. (2014) proposed the development of an antioxidant beverage that could low fat.In addition, experiments have shown that antioxidant activity can be enhanced when a blend of different antioxidant sources is realized (Awe et al., 2013) resulting in better functional qualities of the product.For example, Huang et al. (2009) demonstrated that addition of orange peel extract into black tea resulted in better anti-obesity effect than black tea or orange peel extract alone.
Linear programming is an important tool that is aimed to maximize or minimized a linear function taking account linear equality and/or inequality constraints.Most of the linear programming models have been used in diet optimization, based on the impact of cost and nutrition as constraints (Moraes et al., 2012).However, some research have demonstrated the importance of considering the palatability of the product as a constraint (Darmon et al., 2006).On this background, we have used the sensorial evaluation to define the constraint of a programing model to optimize a drink formulation in dependence of the polyphenol content of cocoa bean shells, coffee silver skin and orange peel.

Extract preparation
Fresh samples of cocoa bean shells, coffee silverskin and orange peel were dried individually at 60 o C in an oven with air circulation during 16 h.The dried residues were mixed according to the formulations obtained from the experimental design; a 10% aqueous extract was prepared by boiling the mixture at 100 o C for 5 min.Higher levels of aqueous extract provided too much astringency to the infusion, resulting in a unpleasant taste.The preparations were allowed to sediment for 5 min at room temperature and the solids were separated by filtration.

Experimental design
An experimental design was obtained applying the equation 1: Here, Yi is the result; the beta symbols (β) represent the coefficients to be adjusted by regression analysis; X 1 , X 12 , Xij 2 are the lineal, interaction and quadratic effect of the components respectively, in which X 1 represent the cocoa bean shells, X 2 represent coffee silverskin and X 3 represent orange peel.The experiments were performed according to the following restrictions: The restrictions of the model were defined based on previous experiments where the combination of those components displayed a pleasing flavor.Due to the restrictions applied, the simplex region was sub divided into a new region in which were included the possible mixture combinations as described by Silveira and Leite (2010).In this case, the design of extreme vertices was applied to the experiments with mixtures.

Dairy drink preparation
Preliminary experimental work was realized in order to define the composition of the drink through trial and error assay.The final acceptable formula was mixing the aqueous extract from by products to skim milk (37%), sugar (5.87%), stabilizers and preservatives (0.13%) and pasteurized at 71.7 o C for 15 seconds, followed by fast cooling at 4 o C for storage and later evaluation.

Sensorial evaluation
The sensorial panel was composed of 30 semi-trained participants who were asked to evaluate taste, odor, color and appearance of the samples.Sensorial evaluations were done applying the test of multiple comparisons (ABNT, 1995).Samples were coded with random numbers of 4 digits and compared to a standard labeled as 'R', a cold commercial drink of mocha coffee with characteristic similar to the drink we aimed to design.Panelists compared each coded sample with the reference sample and graded them using a sensorial scale of 9 points (Table 1).
The data was analyzed applying a variance analysis and Dunnett method for multiple comparisons (Dunnett, 1955).

Total polyphenol content and DPPH assay
Total polyphenol content (TPC) was determined by the Folin-Ciocalteu method and the polyphenol activity was expressed as Gallic Acid Equivalent (GAE) (Lachman et al., 1998).Briefly, 250 µl sample were mixed with 250 µl Folin-Ciocalteu's reagent (Sigma-Aldrich).Na 2 CO 3 20% (750 µl) was added after 5 min and the mixture allowed to proceed for 2 h; absorbance was measured at 765 nm.The antioxidant activity (AA %) of each component was determined by the free DPPH assay with some modifications.The reaction mixture consisted of 0.5 ml sample, 3 ml absolute ethanol and 0.3 ml of a 0.5 mM DPPH in ethanol.Absorbance was measured at 517 nm

Caffeine analysis by HPLC
The assay for caffeine determination was adapted from the method described by Amaro et al., 2008.A Perkin Elmer 200 instrument was used for high pressure liquid chromatography (HPLC), equipped with UV-VIS detector and a C18 column 250 x 4.20 mm with a solid phase of 5 µm.The mobile phase used was a 40:60 mix HPLCgrade methanol-water, with a flow rate of 1ml/min, 5 min ejection time and detection at 254 nm.Injection volume for samples and standards was 10 µl.

Optimization
A lineal programming model was developed with the aim to maximize the total polyphenol content in the formulations, limiting this approach to those formulations that scored similar to the reference in the sensorial evaluation.The target objective function was: Maximize Where Z is the total polyphenol content (TPC) expressed as GAE mg/g in the sample, c j is the TPC (GAE mg/g) measured in ingredient j and x j is the proportion of ingredient j used in the formulation; b j and u j are the minimum and maximum percentage of ingredient j that can be used in the formulation, respectively.

RESULTS AND DISCUSSION
TPC was measured in cocoa bean shells, coffee silverskin and orange peel with the aim to establish the objective function.As shown in Table 2, orange peel had the highest TPC, and was in accordance with the value (23.3 mg GAE/g) reported by Khan et al. (2010).
The major polyphenols present in orange peel are flavanone glycosides, polymethoxylated flavone aglycons, flavone glycosides and C-glycosylated flavones (Sawalha et al., 2009) and health benefit such as anti -inflammatory, anticarcinogenic, anti-viral, antioxidant, anti-thrombogenic and anti-atherogenic properties has been reported it (Middleton et al., 2000;Whitman et al., 2005;Lai et al., 2007;Li et al., 2009).TPC reported for coffee silverskin was higher that the value found in the literature 0.17 mg EGA/g sample (Jiménez-Zamora et al., 2015).Phenolic compounds present in coffee silverskin are similar to the phenolic in coffee brews, and chlorogenic acid is the main polyphenols present (Bresciani et al., 2014).Higher values of TPC were observed in the samples of cocoa bean shells compared to the values (2,56 GAE mg/g) reported by Bruna et al. (2009).
Thirteen formulations were obtained from the design of extreme vertices mixtures (Table 3).The average values for each formulation in each sensorial variable were as shown in Table 3. Taste was rated from 4.90 (Run 2) to 7.23 (Run 8); odor was rated from 6.30 (Run 13) to 7.27 (Run 11); color was rated from 6.13 (Run 13) to 7.77 (Run 9) and a Scores are based on a 9-point scale with 1, extremely inferior to R; 5, no difference; and 9, extremely better than R, b Factors were: X 1 : Cocoa bean shell, X 2 : Coffee silverskin, X 3 : Orange peel, *Mean values are not significantly different from R, c cold commercial drink of mocha coffee (p<0.05) appearance was rated from 6.43 (Run 3 y 13) to 7.87 (Run 8 y 9).The results show that most of formulations were rated as slightly or moderately superior to the reference drink in the variables odor, color and appearance.However, only the formulations 2, 3 and 13 were rated as 'similar to R' regarding taste.The formulations 1, 5, 7, 9, 10 y 12 were rated as 'slightly superior to R' while the formulations 4, 6, 8 and 11 were rated as 'moderately superior to R'.
Because the differences for all 4 sensorial variables were statistically different from the reference R, the formulations 4, 6, 8 and 11 were selected to define the constraints for the analysis of lineal optimization.The objective function equation and the constraints are presented in Table 4.The model was solved using Solver, Microsoft Excel 2010.
The model indicated an optimal formulation of: 74% cocoa bean shells, 24.50% coffee silver skin, and 1.50% orange peel.The predicted TPC for the proposed model was 5.41 mg GAE/g sample, while the observed value was 5.74 ± 0.41 mg GAE/g sample.TPC, antioxidant activity and caffeine content was calculated for the optimal formulation and higher polyphenol, antioxidant activity (82,20 ± 0,08%) and caffeine (114.78 ± 2.83 mg/l) content than the reference 'R' was found (Table 5).Chlorogenic acid present in coffee silverskin may interact with the caffeine and both contribute on fat reduction (Martinez-Saez et al., 2014).Orange peel extract, caffeine and polyphenols present in coffee and cocoa by product extract may exhibit a synergistically effect on anti-obesity (Huang et al., 2009).Furthermore, the levels of caffeine found in the formulation could have positive effects in the improvement on neuromuscular coordination and cognitive function and elevation of mood (Glade, 2010).However, antagonistic interactions may occurs among the caffeine and conjugated linoleic acid present in the milk because of the pro-oxidant properties of caffeine (Anesini et al., 2012).
Several investigations have indicated that adding milk to tea or cocoa products cause negative effect because of an interaction between polyphenols and milk protein reduce the antioxidant activity (Arts et al., 2002 andNiseteo et al., 2012) and cause bioavailability loss (Serafini et al., 1996).However, milk was added into the formulation to lower the astringency cause by the addition of coffee silverskin.A low loss in nutritional quality was obtained considering that only a small part of proteins can interact with the polyphenols present in the matrix (Gallo et al., 2013).Additionally, the addition of milk and sugar may stabilize the antioxidant activity of the beverage (Vasundhara et al., 2008)

CONCLUSIONS
An optimal formulation was obtained by lineal optimization, containing cacao by-products at 74%, coffee silver skin at 24.5% and orange peel at 1.50%.This formulation had 5.74 ± 0.41 GAE mg/g for TPC levels, antioxidant activity by elimination of radicals DPPH 82.20 ± 0.08% and 114.78 mg/l caffeine, demonstrating the feasibility to prepare a dairy drink with acceptable sensorial traits and rich in antioxidant compounds.This values represent a positive predictor for the acceptance of the product in the market.In addition, this model can be used to reformulate the product according to the variation in polyphenol content of raw materials, which are expected to fluctuate since they are generated as residues from the agricultural industry.

Table 5 : Total polyphenols and antioxidant activity
Mean values in the same row not followed by the same letter are significantly different (p<0.05)