Acetylation of Amaranthus viridis starch: Modeling and process parameters optimization

Abstract The optimum reaction conditions for the derivation of acetylated (esterified) starch using response surface methodology (RSM) and artificial neural network (ANN) were studied. All the independent variables (starch solids, acetic anhydride concentration, and reaction time) were of significant (p < .05) importance in achieving esterified starch of Amaranthus viridis. Optimum conditions of 152.46 g of starch, 11 ml of acetic anhydride and time of 2.92 min with corresponding acetyl content and degree of substitution (DS) of 1.74% and 0.06, respectively, were established for ANN. The RSM gave optimum conditions of 149.57 g (starch), 10.38 ml (acetic anhydride) and 3 min (time) with corresponding acetyl content and DS of 1.61% and 0.06, respectively. The order of priority of the process variables was established as acetic anhydride (42.59%), starch solids (33.90%), and reaction time (23.51%). The results provided useful information on development of economic and efficient acetylation process for modification of A. viridis starch.


| INTRODUC TI ON
Genus Amaranthus contains over 60 species but only a few are cultivated, and many are considered weeds (Marin, Narcisa, & Popa, 2008). Amaranth is majorly cultivated for leaf and grains in many temperate and tropical regions. Amaranth is well known around the world and has become established for food use (both grain and leaves) in places like Africa, Central America, Southeast Asia, South America, and North America. The Amaranthus genotype species are cultivated as "pseudo cereals" due to their high content of carbohydrates, proteins and fats, comparable or even superior to cereals (Rusu, Marin, Moraru, Pop, & Cacovean, 2009;Toader & Roman, 2009). A seed of grain amaranth on average contains 13.1-21.0% crude protein; 5.6-10.9% crude fat; and 50-69% starch (Grobelnik-Mlakar, Turinek, Jakop, Bavec, & Bavec, 2009). The average yield per hectare of amaranthus in Nigeria is low (7.60 t/ha) relative to values reported from United States of America (77.27 t/ ha) and the world average (14.27 t/ha) (FAO, 2007). Amaranthus viridis is an underutilized grain with little or no industrial application at present. However, its starch could have industrial applications as cheap alternative source of starch for food industry. Amaranthus viridis is easily cultivated with low labor cost and high grain yield compared to some other sources of starch such as cassava, corn, and breadfruit. Corn, cassava, and potato, which are major sources of starch for food industry, have some other domestic and industrial uses, which posed high demand on them. Moreover, the uses of starch in the food industry are becoming enormous at present. Therefore, there is need for alternative sources of starch from crops of lesser domestic and industrial demand.
Starch is an important food ingredient in the food industry. It is reported that about 53% of starch total production is used in the food sector (sweets -18%, soft drinks -11%, other foods -24%; nonfood sector (total share of 46%); 28% is used for production of paper, cardboard and corrugated board, 13% is used for fermentation, and others 5% (Gérard, Colonna, Buléon, & Planchot, 2001;Hejazi et al., 2008). The uses of starch in food industry include frozen foods, dairy products, soups, sauces, canned foods, beverages, condiments, confectionery and gum, meat products, jams and jellies, syrups and sweeteners, and baking products. However, the limitations of native starch include low shear stress resistance, high retrogradation and syneresis, and poor solubility in common organic solvents (Kavlani, Sharma, & Singh, 2012). Therefore, the functionalities of starch in food industries can be enhanced through modifications. One of such modification techniques is acetylation (esterification). Esterification (often called acetylation) is a chemical modification mechanism, in which the hydroxyl groups are replaced with acetyl groups thereby leading to steric obstacles and a subsequent decrease in the gelatinization temperature. The acetylation reduces retrogradation and improves the stability at cooling and freezing points (Kavlani et al., 2012).
The starch properties of the seeds of some amaranth cultivars have been characterized (Baker & Rayas-Duarte, 1998;Hoover, Sinnott, & Perera, 1998;Radosavljevic, Jane, & Johnson, 1998;Marcone, 2001;Choi, Kim & Shin, 2004). However, there is a dearth of information on modification of A. viridis starch, using the esterification (acetylation) technique. It is worthy of mentioning that the Design of Experiments (DOE) approach was not used in most of these reports on Amaranthus starch.
Modeling and optimization of processes involved in the food processing industry can be used to improve the yield of the target products. Rather than the typical one-factor-at-a-time method of optimization, which does not describe the complete effects of the variables in the process and does not consider the interactions between the variables, Response Surface Methodology (RSM), which defines the effect of the independent variables, alone or in combination in a process, is nowadays being applied in modeling and optimization studies (Bas & Boyaci, 2007;Betiku & Taiwo, 2015). This tool has been widely used in many areas of food research, such as production of dairy tofu (Chen, Chen, & Lin, 2005), ethanol production (Betiku & Taiwo, 2015), citric acid production (Betiku & Adesina, 2013;Dhillon, Brar, Verma, & Tyagi, 2011), lactic acid production (Naveena, Altaf, Bhadraya, Madhavenda, & Reddy, 2005) and oxalic acid production (Emeko, Olugbogi, & Betiku, 2015). Artificial Neural Network (ANN), which is a computational method that can mimics the neurological processing capability of the human brain, has also been applied to modeling of many food processing studies. These studies include gluconic acid (Osunkanmibi, Olowlabi & Betiku, 2015), ethanol (Betiku & Taiwo, 2015) and oxalic acid (Emeko et al., 2015) production processes as well as in enzymatic reaction catalyzed by amyloglucosidase (Bas & Boyaci, 2007). Many of these studies have demonstrated consistently that the predictive capability of ANN is stronger than RSM (Bas & Boyaci, 2007;Betiku & Taiwo, 2015;Emeko et al., 2015). Therefore, this work aims to investigate the acetylation of A. viridis starch with the view to enhancing the utilization of the starch in food industry. The acetylation process was modeled, using both RSM and ANN. The vital variables of the acetylation process investigated include starch solids, acetic anhydride concentration, and the reaction time. The variables were optimized, using RSM and ANN coupled with genetic algorithm.

| Materials
Amaranthus viridis grains were cultivated by National Horticultural Research Institute (NIHORT), Kano, Nigeria and matured grains were collected after 14 weeks (April to July, 2015). The grains were wet cleaned and dried in a hot air oven (SM9053, Uniscope, UK) at 50°C for 8 hr. All chemicals were of analytical grades and obtained from Fisher Scientific (Oakville, ON, Canada) and Sigma Chemicals (St. Louis, MO, USA).

| Isolation of starch from amaranth grains
Starch was extracted as described by Kong, Bao, and Corke (2009) with some modifications. The amaranth grains were soaked in distilled water (1:5, w/v) maintained at 28 ± 2°C for 12 hr. The seeds F I G U R E 1 Flow chart for extraction and acetylation of Amaranthus viridis starch were rinsed, drained, and wet milled in attrition mill (Double Win, FS450, China). The resultant slurry was filtered through 149 μm mesh sieve. The filtrate was then centrifuged at 4552 × g, using a centrifuge (0502-1, Hospibrand, USA) for 20 min. The starch layer was redispersed in distilled water (1:5 w/v), and centrifuged as earlier described, and this procedure was carried out in duplicate. The isolated starch was dried in an air oven (Uniscope, SM9053) at 50°C for 36 hr and ground, using a hammer mill to pass through a sieve with mesh sieve 212 μm. The flow chart for the process is as shown in Figure 1.

| Experimental design and RSM modeling
The orthogonal central composite design (CCD) of RSM was used for this work. The three independent variables considered for the modeling include starch solid (50-150 g), acetic anhydride (5-15 ml) and time (3-13 min). The number of experimental conditions (Table 1) used for this work was generated using Equation (1), distance of the star points from the center point (2); real values of the center and star points were obtained, using Equations (3) and (4), respectively.
The order of experimentation was completely randomized to avoid systematic errors.
where, k is the number of independent variables, n o is the number of centre point and 2k is the number of star point, X high level is value of independent variable at high level, X low level is value of independent variable at low level, X mean is mean of values of independent variable at low level and high level, X range, is difference between values of independent variable at high level and low level.
The following statistical indicators were employed: coefficient of determination (R 2 ), adjusted (Adj. R 2 ), probability value at 95% confidence interval, predicted R 2 , coefficient of variation, lack-of-fit, and analysis of variance (ANOVA). The modeling and optimization of the esterification process were carried out, using RSM of the Design Expert software version 8.0.7.1 (Stat-ease Inc., MN, USA). Pareto chart was developed using Statistica software, version 12.0 (Stat Soft, Inc., 2014).

| Derivation of acetylated starch and process parameters optimization
The acetylation process was carried out as described by Lawal (2004) with some modifications. The extracted native amaranth TA B L E 1 Acetyl content and degree of substitution of acetylated starch starch varying from 50 to 150 g (Table 1) was dispersed in 500 ml of distilled water and magnetically stirred for 20 min. The pH of the slurry was adjusted to 8.0. Acetic anhydride varying from 5-15 ml was added while maintaining a pH range of 8.0-8.5. The reaction was allowed to proceed for time varying from 3 to 13 min after the addition of acetic anhydride. Thereafter, the pH of the resultant slurry was adjusted to 7 using 0.1 mol/L HCl solution, centrifuged at 4552 × g, using a centrifuge (0502-1 Hospibrand)

Degree of substitution
for 15 min, washed with distilled water (1:10 w/v) four times, dried in a hot air oven (Uniscope, SM9053) for 36 hr, milled using attrition mill, sieved with mesh 212 μm and packaged in an airtight plastic container. The flow chart for the process is shown in Figure 1.

| Determination of percentage acetylation and degree of substitution
The methods described by Medinav, Pardo, and Ortiz (2012)  where, blank is ml of HCl used in the native starch titration, sample is the ml of HCl used in the esterified starch titration, 0.043 is milliequivalents of the acetyl group, 162 is the molecular weight of glucose, 4300 is molecular weight of the acetyl group multiplied by 100, and 42 is the molecular weight of the acetyl group −1.

| Modeling and optimization using ANN
A commercial software, NeuralPower version 2.5 (CPC-X Software), was used for the ANN modeling and optimization. The dependent variables (%acetyl and DS) were predicted by using multilayer full feed forward (MFFF) and multilayer normal feed forward (MNFF) neural networks, which were trained by different learning algorithms such as incremental back propagation (IBP), quickprob (QP), genetic algorithm (GA), batch back propagation (BBP), and Levenberg-Marquardt algorithm (LM). Each ANN was trained using a default stopping criteria of 100,000 iterations. The dataset in Table 1  where, n is the number of experimental data, a p,i is the predicted values, a p,ave is the average predicted values a e,i is the experimental value, a e,ave is the average experimental values, and j is the number of input variables (Akanbi, Adeyemi, & Ojo, 2006;Emeko et al., 2015).

| Acetyl content and degree of substitution
The results of the acetyl content and degree of substitution (DS) are presented in the maximum acetyl group allowed in food is 2.50%. The threefree hydroxyl (OH) groups located at C 2 , C 3, and C 6 have different reactivity. The primary OH attached to C 6 is more reactive and is acetylated more readily than the secondary ones at C 2 and C 3 due to steric hindrance and their affinity for OH groups on the neighboring glucose unit (Miyazaki, Hung, Maeda, & Morita, 2006).

| Modeling and parameters optimization of starch acetylation process by RSM
In order to describe the relationship between the dependent variables (acetyl content and DS) and the independent variables (starch solids, acetic anhydride, and reaction time), the dependent variables were fitted by second-order polynomial quadratic regression models. By applying multiple regression analysis on the experimental data obtained for the dependent variables (Table 1), the analysis of variance (ANOVA) generated for the models are presented in Table 2. According to Myers and Montgomery (2002) and Fristak, Remenarova, and Lesny (2012), a large F-value indicates that most of the variations could be explained by the regression equation, whereas a low p-value (p < .05) indicates that the model is considered to be statistically significant.
The fitness and adequacy of the models were judged by the coefficient of R 2 and significance of lack-of-fit. The R 2 is defined as the ratio of the explained variation to the total variation, a measure of than the R 2 (Chan et al., 2009;Myers & Montgomery, 2002). Also, the absence of any lack-of-fit (p > .05) strengthened the reliability of a model.
In the case of acetyl content ( The coefficient of variation, CV, which is independent of the unit is defined as the ratio of the standard deviation of estimate to the mean values of the observed response. The CV is a measure of reproducibility and repeatability of the model (Chen, Xiong, Peng, & Chen, 2010;Chen et al., 2011;Pishgar-Komleh et al., 2012). The CV value of 5.14% observed in this work suggests that the model could be considered reasonably reproducible (CV < 10%) (Chen et al., 2011).
The adequate prediction (Ad. Pred.) compares the range of the predicted value at the design points to the average prediction error.
Adequate prediction measures signal to noise ratio. A ratio greater than 4 is desirable (Rajmohan & Palanikumar, 2013). The value of the adequate prediction for the acetyl content was significantly greater than 4 (39.56). The lack-of-fit test value of 0.3296 was not significant (p > .05), which also shows a good fit between experimental data and the model.  (11) and (12).

| Modeling and parameters optimization of starch acetylation process by ANN
In the derivation of esterified starch process, many neural network architectures and topologies for the estimation and prediction of the dependent variables were tested. The ANN models in the optimal region are presented in Table 3(b). As reported by Betiku and Taiwo (2015), there are many learning algorithm types reported in the literature, thus, it is difficult to know in advance which of the learning algorithms will be more efficient for a given study. According to Betiku and Taiwo (2015), the transfer function types employed affect the neural network learning and aid its performance. Thus, several ANN learning algorithms and transfer functions effects were evaluated by successful training of the neural network models. The results obtained indicated that IBP was the most suitable learning algorithm for the dependent variables synthesis (Table 3b)    esterified starch was acetic anhydride (42.59%), followed by starch solids (33.90%) and then reaction time (23.51%) (Figure 3c).

| Performance evaluation of ANN and RSM for the acetylation process
The extent of accuracy of the developed models from RSM and ANN were examined using R 2 and E ( Table 3). The average R 2 of RSM and ANN were .9749 and .9928, respectively; and E values of 0.09 and 0.05% for RSM and ANN, respectively. Thus, the ANN model proved to be more effective due to the higher value of R 2 and lower value of E.
Similar observations were reported in the modeling and optimization studies on enzymatic reaction catalyzed by amyloglucosidase (Bas & Boyaci, 2007), ethanol production from breadfruit starch (Betiku & Taiwo, 2015) and oxalic acid production from cashew juice (Emeko et al., 2015). Conclusively, ANN performed better than RSM in the modeling and optimization of acetylation process for A. viridis starch.

| CON CLUS ION
The work examined the acetylation (

CO N FLI C T S O F I NTE R E S T
The authors declare no conflicts of interest.

ACK N OWLED G M ENTS
The authors thank the the Department of Food Science and