- Email: [email protected]
Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties
Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties Tom O. Jondiko, Liyi Yang, Dirk B. Hays, Amir M.H. Ibrahim, Michael Tilley, Joseph M. Awika PII: DOI: Reference:
S1466-8564(16)00016-3 doi: 10.1016/j.ifset.2016.01.010 INNFOO 1464
To appear in:
Innovative Food Science and Emerging Technologies
Received date: Revised date: Accepted date:
12 October 2015 31 December 2015 1 January 2016
Please cite this article as: Jondiko, T.O., Yang, L., Hays, D.B., Ibrahim, A.M.H., Tilley, M. & Awika, J.M., Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties, Innovative Food Science and Emerging Technologies (2016), doi: 10.1016/j.ifset.2016.01.010
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties
RI
PT
Tom O. Jondikoa,b, Liyi Yanga,b, Dirk B. Haysa, Amir M.H. Ibrahima, Michael Tilleyc, and Joseph M. Awikaa,b,*
Department of Soil and Crop Sciences, Texas A&M University, College Station, TX, 77843
b
Department of Nutrition and Food Science, Texas A&M University, College Station, TX, 77843
c
USDA-ARS Center for Grain and Animal Heatlh Research, 1515 College Avenue, Manhattan, KS 66502, USA
MA
NU
SC
a
AC CE P
TE
D
*Corresponding author: Email: [email protected]; Phone: 979-845-2985, Fax: +1 979-845-0456.
The use of trade, firm, or corporation names in this presentation is for the information and convenience of the reader. Such use does not constitute an official endorsement or approval by the United States Department of Agriculture or the Agricultural Research Service of any product or service to the exclusion of others that may be suitable.
1
ACCEPTED MANUSCRIPT Abstract
Traditional wheat quality methods for bread have poor predictive power for flatbreads quality,
PT
which impedes genetic improvement of wheat for the growing market. We used a multivariate discriminant analysis to predict tortilla quality using a set of 16 variables derived from kernel
RI
properties, flour composition, and dough rheological properties of 187 experimental hard wheat
SC
samples grown across Texas. A discriminant rule (suitability for tortillas = diameter > 165mm +
NU
day 16 flexibility score >3.0) was used to classify samples. Multivariate normal distribution of the data was established (Shapiro-Wilk p > 0.05). Logistic regression and stepwise variable
MA
selection identified an optimum model comprising kernel weight, glutenin-gliadin ratio, insoluble polymeric proteins, and dough extensibility and stress relaxation parameters, as the
D
most important variables. Cross-validation indicated 83% prediction efficiency for the model.
TE
This work provides important insight on potential targets for wheat quality genetic improvement
AC CE P
for tortillas and specialty products market.
Keywords Wheat quality; tortilla; flatbread; gluten; genetic improvement
2
ACCEPTED MANUSCRIPT 1. Introduction Tortillas and other specialty breads are increasingly becoming a global dietary staple. For
PT
example, according to the Tortilla Industry Association report for 2013, wheat tortilla was the
RI
only bakery segment that experienced growth in 2012 and is projected to increase further (TIA, 2013). Wheat tortilla sales exceeded $ 6 billion in 2012, affirming consumer preference for its
SC
versatility and functional convenience. To consumers, the definition of good quality tortilla
NU
encompasses its ability to retain flexibility and be large enough to wrap food (Waniska, et al., 2004).
MA
Despite the growing popularity of tortillas, the main challenge is that there is no reliable and practical method to predict wheat (Triticum aestivum L.) functionality for tortillas, as is the
TE
D
case with pan bread and other mainstream baked products. Currently, most tortilla ingredient suppliers and processors use trial and error and additives to optimize tortilla quality, which
AC CE P
compromise sensory appeal and quality, and adds to cost of manufacture. The health conscious consumer is also increasingly demanding use of fewer additives (clean label), as well as healthier options, like whole grains. Thus there is a need to develop wheats with improved functionality for the tortilla and flatbreads market. However, effective genetic improvement efforts require that the key grain quality parameters responsible for desirable product characteristics can be reliably identified. Ideally the methods to identify these grain quality factors should be rapid and require small sample size. Currently, the only way to predict wheat functionality for tortillas is to actually make the product. Tortilla quality in the US is largely defined by diameter, flexibility during storage, and opacity (Alviola & Awika, 2010; Alviola, Jondiko, & Awika, 2010). Earlier studies reported that tortillas of good quality can be produced using wheat flour of intermediate protein content, 3
ACCEPTED MANUSCRIPT protein strength and low level of starch damage by (Waniska, et al., 2004). However, the vast majority of wheat that fall in this category did not produce a good quality product (Waniska, et
PT
al., 2004). Tortilla diameter can be predicted using linear equations comprising mixing time and
RI
dough resistance to extension (Barros, et al., 2010). However, these linear models did not significantly predict tortilla flexibility, which is a critical tortilla quality attribute. A major
SC
challenge to predicting tortilla quality is the negative correlation that generally exists between
NU
the main quality factors, tortilla diameter, and flexibility during storage (Mondal, et al., 2008; Pascut, Kelekci, & Waniska, 2004). Large diameter generally requires weak and extensible
MA
dough, which tends to be detrimental to tortilla flexibility. Genetic improvement of wheat targeting alterations of high molecular weight glutenin
TE
D
sub-units (HMW-GS) composition through deletion at one or more of the Glu A1, B1, and D1 loci has shown promise as a way to produce unique gluten functionality ideal for tortillas and
AC CE P
other specialty products (Jondiko, et al., 2012; Mondal, et al., 2008; Tuncil, et al., 2016; Zhang, et al., 2014). There is a need to develop prediction models that can be used to screen these wheat varieties under development for end use functionality. Hence, this study used multivariate statistical methods designed to elicit information from simultaneous measurements of many variables acquired from wheat kernel, flour, and dough to predict the quality of tortilla, specifically the diameter and flexibility during storage. In addition, the study investigated whether these multivariate models could be used to reliably classify early to late generation wheat lines for their potential to produce good quality tortillas. The objectives of this study were to develop a model that can be used to predict the functional performance of wheat varieties for tortilla processing, and provide insight on wheat quality parameters that can be targeted for genetic improvement for specialty flatbreads market. 4
ACCEPTED MANUSCRIPT
2. Materials and Methods
PT
2.1. Experimental materials
RI
A total of 187 hard winter wheat (HWW) lines were used. These included advanced generation breeding lines comprising 38 Texas elite (TXE) lines, 3 uniform variety trial (UVT)
SC
lines; 101 experimental lines derived from ‘TAM111’/‘TAM112’ population (TAM1112); and
NU
45 identity preserved lines specifically developed for specialty flat breads (TIA). These TIA lines were selected based on variations in their allelic composition at the HMW-GS loci Glu A1, Glu
TE
2.2. Kernel properties and milling
D
across Texas in 2009 – 2012 seasons.
MA
B1 and Glu D1. The wheat samples were grown in Texas A&M AgriLife experimental plots
The wheat lines were evaluated for hardness, diameter, weight and moisture content
AC CE P
using a single kernel hardness tester (SKCS) (Perten Instruments, Springfield, IL). The grains were tempered to 14% moisture content (AACC Method 26-50.01) and milled using a Quadrumat Senior mill (Brabender Instruments, South Hackensack, NJ). Flour milling yields were recorded. Samples were processed into tortillas with 1- 2 months after milling. 2.3. Evaluation of flour properties 2.3.1. Total protein content. Total flour protein content was determined in two replicates for each wheat line using near-infrared reflectance spectroscopy (Perten PDA 7000 Dual Array with Grams Software) according to AACC Method 39-11.01 (AACC-International, 2010).
2.3.2. Polymeric to monomeric protein ratio (glutenin to gliadin ratio). Protein extraction of proteins followed the method of Gupta, Khan, & Macritchie (1993). Briefly, a 10 mg flour sample were mixed with 1 mL 0.05 M Sodium phosphate buffer, pH 6.9, containing 0.5% SDS 5
ACCEPTED MANUSCRIPT (w/v) then sonicated for 35 s at power setting 10 W. The sample was then be centrifuged at 15,000 × g for 5 min and the supernatant collected (contains total protein) and filtered through
PT
0.45 μm filter and analyzed by size – exclusion HPLC using an Agilent 1260 HPLC (Agilent,
RI
Santa Clara, CA), 300 x 7.8 mm BioSep-SEC-S400 column (Phenomenex, Torrance, CA) with
gradient system composed of 50% ACN+0.1% TFA (B) and 50% water+ 0.1% TFA (A), 30°C
SC
column temp., at a flow rate of 1 ml/min for 30 min run. The chromatograms were manually
NU
integrated. The area of the first peak corresponds to total polymeric proteins and the area of the second peak to monomeric proteins (Gupta, Khan, & Macritchie, 1993). Two replicates of each
MA
flour sample were analyzed.
D
2.3.3. Insoluble polymeric protein content (IPP). Ten mg of flour was suspended in 1 mL of 0.05
TE
M sodium phosphate buffer (pH 6.9), containing 0.5% sodium dodecyl sulfate (SDS) and shaken on a vortex for 30 min. The mixture then centrifuged for 5 min at 16,000 × g. The supernatant
AC CE P
(containing soluble polymeric protein - SPP) were collected and filtered (0.45 μm) and analyzed by size – exclusion HPLC as described above (Bean, et al., 1998). The pellet were mixed with 1 ml sodium phosphate buffer and sonicated for 25 s at 10 W. The mixture were centrifuged at 16,000 × g /5 min, and the supernatant collected and filtered as above then analyzed using the SE-HPLC as described above. The percentages of soluble (extractable) and insoluble (unextractable) polymeric protein were calculated as [peak 1 area (extractable)/peak 1 area (total)] × 100 and [peak 1 area (unextractable)/peak 1 area (total)] × 100 respectively. Peak 1 (total) refers to the sum of peak 1 (extractable) and peak 1 (unextractable). 2.3.4. High molecular weight to low molecular weight glutenin sub-units ratio (H_L_GS_Ratio). HMW-GS and LMW-GS were quantified using RP-HPLC. A sample of 100 mg flour was mixed with 1 mL sodium iodate buffer (0.3 M sodium iodate + 7.5% isopropanol) and vortexed for 15 6
ACCEPTED MANUSCRIPT min. The mixture was centrifuged for 5 min at 15,000 × g. The supernatant containing gliadins were discarded. To the pellet 1ml water were added then shaken for 5 min and centrifuged at
PT
15,000 x g for 5 min. The pellet were mixed with 1 mL 50% isopropanol containing 2% BME and vortex for 30 min, and then centrifuged for 5 min. at 15,000 × g. The supernatant was
SC
RI
collected (contains glutenins) and 600 L of the glutenin extract was alkylated with 40 µL 4vinylpyridine for 15 min at 60°C. The resulting sample was injected into a Phenomenex Jupiter
NU
C18 250 × 4.6 mm diameter column, 5 μ particle size, and 300 Å pore size. The solvent flow rate was 1.0 mL/min and composed of water (A) and acetonitrile (B), both containing 0.1% TFA.
MA
The gradient was as follows: 0-3 min from 25% B to 35% B, 3-24 min increased to 53%B, the gradient decreased to 25% B at 25 min and kept at 25% B until 29 min. Detection of protein
TE
D
peaks were carried out by UV detector at 210 nm. The area of the curve corresponding to HMWGS and LMW-GS contents were determined by manual integration and the HMW/ LMW-GS
Fu, 2003).
AC CE P
ratio calculated (Cinco-Moroyoqui & MacRitchie, 2008; Fu & Kovacs, 1999; Suchy, Lukow, &
2.4. Dough mixing and rheology
2.4.1. Dough development time. A Mixograph (National Manufacturing Co., Lincoln, NE) was used to estimate dough mixing/ development time and flour water absorption; ten grams of flour were used (14% mb) (AACCI Method 54-40.02; AACCI 2010). Mixing/ dough development time was manually calculated from the mixograph by drawing two midlines from each end of the graph. The point of crossover was marked as the peak time for each wheat line (Alviola & Awika, 2010).
7
ACCEPTED MANUSCRIPT 2.4.2. Dough compression force. Dough compression test was used to measure the maximum compression force required to deform a 45 g dough ball. Two dough balls were compressed to
PT
70% of their height using 10 cm diameter probe on a TA-XT2 texture analyzer (Texture
RI
Technologies Corp., Scarsdale, NY). Maximum dough compression force was calculated as an
SC
average force of the two dough balls for each of the wheat lines. (Bejosano, et al., 2005)
NU
2.4.3. Stress Relaxation test. Stress relaxation tests were conducted by compressing two 45 g
MA
dough balls on a TA-XT2 texture Analyzer. This was conducted on dough balls that had been rested for10 min. A 10 cm diameter cylindrical probe was
lowered at 1 mm/s on each dough
D
ball and held for 120 s. The relaxation force at 25 s, 100 s, maximum force, and relaxation time
TE
were recorded using the texture expert software (Jondiko, et al., 2012).
AC CE P
2.4.4. Dough extensibility evaluation. Dough extensibility properties: elasticity (N), extensibility (mm) and work to extend (N.mm) were determined using the modified dough extensibility test (Jondiko, et al., 2012). Doughs were prepared using 100 g flour and 2 g salt for each of the wheat lines and mixed in a Mixograph 100 g micro-mixer. Warm water (~ 35º C) was added to the dry ingredients based on an adjusted mixograph water absorption value and mixed to optimum dough development time, based on the mixograph test (Jondiko, et al., 2012). Dough extensibility parameters were evaluated on 10 dough strips from each wheat line as previously described (Smewing, 1995).
2.5. Tortilla formulation and processing
8
ACCEPTED MANUSCRIPT Tortilla formulation included two batches of 500 g flour from each of the wheat lines, without reducing agents as described (Alviola, Jondiko, & Awika, 2012). The quantity of
PT
distilled water used was based on mixograph water absorption for each line, minus 10% points.
RI
Mixed dough was divided and rounded into approx. 45 g dough balls (Duchess Divider/Rounder, Bakery Equipment and Service Co., San Antonio, TX). The dough balls were rested at 32º C and
SC
65-70% relative humidity in the proofing chamber for 10 minutes, and then hot-pressed (205°C,
NU
7.9 kPa, 1.4 sec) into round discs (Bello, et al., 1991) and baked for 30 sec at 190 and 200 °C in a gas fired three-tier oven (Model 0P01004-02, Lawrence Equipment, El Monte, CA). Tortillas
MA
were then cooled to room temp and packed in 1 mm thick polyethylene bags and stored at 22oC
TE
D
for textural and shelf stability evaluation.
AC CE P
2.6. Tortilla quality evaluation
2.6.1. Tortilla physical property evaluation. Tortilla physical properties: weight, diameter, and thickness were evaluated on ten randomly selected tortillas from each of the wheat lines following the method of Friend and others (Friend, et al., 1995 ). These properties were used to determine the tortilla specific volume as described below: ). Where; height = height of a single tortilla (cm); weight = weight of a single tortilla (g), r = average radius of a tortilla (cm) (Waniska, et al., 2004). Tortilla lightness (L* – values) was measured on two randomly selected tortillas at two different points of each side of the tortillas using a Minolta color meter (Chroma Meter CR-310, Minolta, Tokyo, Japan). 9
ACCEPTED MANUSCRIPT 2.6.2. Subjective tortilla shelf stability evaluation. Tortilla rollability/ flexibility score was determined after 16 days of storage, using procedures recently described (Alviola & Waniska,
RI
indicative of inferior/ poor shelf stability (Alviola, et al., 2008).
PT
2008; Jondiko, et al., 2012). Flexibility scores < 3 (numerous cracks and breaks on tortillas) was
SC
2.6.3. Objective tortilla shelf stability evaluation. The two-dimensional extensibility test (Bejosano, et al., 2005) was used to measure tortilla textural changes after day 0, 4, 8, 12 and 16
NU
of storage. The extensibility test was determined on four tortillas from each of the wheat lines at
MA
each time point using a TA-XT2i texture analyzer. The analyzer settings were: return to start option, trigger force (0.05 N), pre and post-test speeds (10.0 mm/s) and test speed (1.0 mm/s).
D
The texture expert software was used to record tortilla texture properties: Gradient/ modulus of
AC CE P
TE
deformation (N/mm), force (N), distance (mm) and work to rupture (N.mm).
2.7. Statistical analysis
2.7.1. Data description and classification criterion. A total of 16 continuous variables were measured on the 187 wheat lines using the methods described above. The 45 wheat lines possessing variations in high molecular weight glutenin sub-units (HMW-GS) at the Glu A1, Glu B1 and Glu D1 were used to train and calibrate the prediction models. This was based on the evidence that these lines had the unique protein functionality to produce tortillas of superior quality (Jondiko, et al., 2012). Sixteen variables measured on all the 187 lines were then used to cross validate the prediction models. These variables were kernel properties (hardness, diameter, and weight); flour properties (protein content, mixograph water absorption, glutenin to gliadin ratio (GGRatio), high molecular weight to low molecular weight glutenin sub-unit ratio 10
ACCEPTED MANUSCRIPT (H_L_GS_Ratio), insoluble polymeric protein content (IPP)); and dough rheology (dough development/ peak time (Mixo_Time), dough compression force (CompForce), resistance to
PT
extension (ForceD), extensibility (DistanceD), work to extend (WORK), stress relaxation time
RI
(RT), stress relaxation force after 25 s (F_25) and after 100 s (F100) of compression). Predication models were developed for both tortilla diameter and day 16 flexibility scores using
SC
the 16 variables measured on wheat kernel, flour and dough rheology.
NU
A suitability of categorical numbers 1 or 0 was assigned to each wheat line, where varieties with “1” were suitable to make quality tortillas (Diameter ≥ 165 cm and day 16
MA
flexibility score ≥ 3.0). Wheat lines not meeting this description were assigned “0” meaning they
D
were not suitable to make quality tortillas.
TE
2.7.2. Test for data normality. Multivariate normal distribution of the data set was determined
AC CE P
using the Shapiro-Wilk statistic p – value for each of the 23 variables (p < 0.05). 2.7.3. Discriminant analysis using logistic regression. Discriminant analysis was done using SAS version 9.2 (SAS Institute, Cary, NC). The data was standardized to eliminate the effect of differences in units of each variable. In order to predict wheat lines suitable to make good quality tortillas, a discriminant analysis were carried out using all the wheat kernel, flour, and dough property variables. Prediction models were developed based on three methods of variable selection. These were forward, backward, and stepwise variable selection. Each model was used to classify all the wheat lines into two suitability classes based on a discriminant rule where if the response (Suitability) based on the variables in each model was equal to one (= 1) for a line that was good for tortilla processing , otherwise the wheat line had inferior tortilla processing ability (response was 0). The models were used to cross validate prediction accuracy using five nearest 11
ACCEPTED MANUSCRIPT neighbors in logistic regression discriminant analysis. An apparent error rate of classification was calculated by dividing the number of lines that were misclassified with the total number of
RI
PT
lines evaluated during cross validation.
SC
3. Results and discussion
NU
3.1. Data distribution and normality
The means, minimum and maximum values for kernel, flour, dough and tortilla
MA
properties for the wheat lines are given in Table 1. Two of the 187 wheat lines were identified as
D
outliers using principal component analysis and eliminated hence, the total number of
TE
observations were N = 185 for all the 16 variables (Table 1). To eliminate the effect of differences in units in the variables, data analysis was conducted on standardized variables. The
AC CE P
variables were standardized using a fixed mean and variance. Kernel property values had a broad range (Table 1) and followed a multivariate normal distribution (p < 0.05) indicating good diversity in the wheat lines. The mean hardness for the TIA, TXE/UVT and TAM1112 lines ranged between 38 and 89. Lines from the TXE/UVT populations had the highest average hardness index (80.3, vs 66.7 for TIA and 59.6 for TAM1112); this was expected because these are advanced materials that have gone through extensive selection towards potential release as hard wheat varieties. Flour protein content (as is) ranged between 11.9 and 16.4%, whereas mixograph mixing time ranged between 1.5 and 7.5 min, with no obvious differences among the different wheat populations. These ranges were broad enough to capture expected variability in wheat quality. Other traits of interest, like IPP content and dough extensibility, that have been shown to influence tortilla quality (Jondiko, et 12
ACCEPTED MANUSCRIPT al., 2012), also had wide ranges (Table 1). Notably, dough extensibility, which correlates positively with tortilla diameter (Jondiko, et al., 2012; Pierucci, et al., 2009) was on average
PT
highest for the TIA lines (82.9 mm, vs 71.1 for TXE/UVT and 55.9 mm for TAM 1112). The
RI
high extensibility of these TIA lines has been previously demonstrated, and is related to the alteration in their HMGS alleles through specific deletions at Glu A1, B1 or D1 genome
SC
(Jondiko, et al., 2012).
NU
Tortilla quality attributes also had a broad range of distribution, representing very poor to
MA
excellent product quality. For example, diameter values ranged from 136 – 184 mm, and day 16 flexibility scores ranged from 1.0 (poor) to 5.0 (excellent) (Table 1). A diameter value of 165
D
mm and a day 16 flexibility score of 3.0 are the minimum values for defining good quality
TE
tortillas based on the protocols used in this study (Bejosano, et al., 2005). Tortilla specific volume (an indicator of leavening effect) ranged between 0.8 and 2.1 cm3/g, whereas lightness
AC CE P
(L value) ranged from 72.6 – 85.6; the L-value is a good indicator of tortilla opacity, with higher values being preferred by USA consumers (Alviola & Awika, 2010; Alviola, et al., 2012). Among the different wheat populations used, the TIA samples had higher average diameter and day 16 flexibility score than the TXE-UVT and the TAM1112 lines (Table 1), which indicates improved functionality of the TIA lines for tortilla production. Diameter typically negatively correlates with flexibility score for tortillas (Pascut, et al., 2004). All tortilla quality parameters were normally distributed (p > 0.05).
3.2. Discriminant analysis and model selection
13
ACCEPTED MANUSCRIPT Variable selection was carried out using forward, backward and stepwise variable selection resulted into three models (Table 2 – 4). Forward selection procedure resulted in a
PT
model with 11 of the 16 variables (Table 2), a total of 149 wheat lines of the 185 lines were used
RI
in this model because 36 lines were eliminated for being outliers based on the variables selected
SC
for the model (Table 5).
Predicted suitability = -1.584 -0.05*Hardness + 9.734*Diameter -0.496*Weight -
NU
1.137*Protein + 7.274*GGRatio - 9.2731*H_L_GS_Ratio -0.053*IPP +1.5851*ForceD -
MA
0.171*Area + 1.354*F_25 + 0.029*CompForce (Table 2). The three kernel properties (hardness, diameter and weight), four flour protein variables
D
(protein, GGRatio, H_L_GS_Ratio and IPP) and four dough rheology variables (extensibility,
TE
work to extend, relaxation force after 25 sec and compression force) were used to classify and
AC CE P
cross validate the model using five nearest neighbors. The apparent error rate for this model was 0.30 (Table 5). This model correctly classified 67% and 72% of the Good and Poor wheat lines with an overall accuracy of 70%. The model was not ideal because it included a lot of variables that are related. For example, both kernel diameter and hardness contribute to kernel weight, whereas dough resistance to extension (ForceD) is a component of work to extend (Area). Backward elimination procedure resulted in the selection of 9 variables for the prediction model (Table 3). One hundred and forty nine lines were used in validating this model after elimination of outliers based on the variables selected. The backward elimination model was as follows:
14
ACCEPTED MANUSCRIPT Predicted suitability = -5.3 -0.052*Hardness + 9.92*Diameter -0.479*Weight 1.075*Protein + 7.846*GGRatio -9.976*H_L_GS_Ratio -0.185*Area + 1.41*F_25 +
PT
0.028*CompForce (Table 3).
RI
This model had a slightly lower apparent error rate (0.26) compared to the forward
SC
selection model. It misclassified 31 % and 22 % of the good and poor wheat lines respectively. It correctly classified 74 % of the wheat lines (Table 5). The backward elimination procedure
NU
resulted in more efficient selection of variables compared to forward selection procedure; it
MA
resulted in better prediction power with fewer variables. It also eliminated one of the redundant variables, ForceD.
D
Stepwise variable selection model had the fewest variables, seven. A total of 127 lines
AC CE P
5):
TE
were used to validate this model after elimination of outliers based on the seven variables (Table
Predicted suitability = - 10.149 -0.189*Weight + 10.583*GGRatio - -0.052*IPP 0.219*Area + 1.401*F_25 + 0.355*F_100 + 0.021*CompForce (Table 4). This model had one variable each from the kernel properties (Weight), dough extensibility test (Work to extend), and compression test (Compression force) plus two variables from both the flour protein fractions (GGRatio and IPP) and stress relaxation tests (F_25 and F_100). The stepwise model had a significantly lower apparent error rate (0.17) compared to forward and backward elimination models. It correctly classified 83% of the wheat varieties based on their tortilla processing functionality (Table 5). This model was the most efficient of the three, it eliminated most of the correlated variables (grain hardness, and diameter; dough resistance to extension), while providing the best prediction power. 15
ACCEPTED MANUSCRIPT In summary, the forward selection model was the least efficient with an apparent error rate of 0.30. This model requires 11 variables which include several related (thus potentially
PT
redundant) variables. Backward elimination model had nine variables and apparent error rate of
RI
0.26. It misclassified 38 of the wheat varieties, whereas forward selection model misclassified 45. Compared to the forward selection model, it was somewhat more efficient, though it
SC
excluded one of the variables that has been consistently reported to impact tortilla quality
NU
parameters, the IPP, while retaining several correlated variable (kernel hardness, diameter, and weight). On the other hand, it eliminated one of the redundant variables, dough resistance to
MA
extension (ForceD). Thus for practical purposes, its advantage over the forward selection model would be marginal. The stepwise model was the most efficient with an apparent error rate of 0.17
TE
D
(83% prediction accuracy), and the fewest number of variables (seven). It misclassified 21 wheat samples. It also excluded all of the redundant variables that were included in the other two
AC CE P
procedures, while retaining key variables like IPP and work to extend dough. This model provides a good foundation for identifying key wheat quality parameters for tortillas, and predicting the suitability of a wheat line for tortilla production. It is worth noting that five variables showed up in all the three models including, grain weight, flour glutenin-gliadin ration, and three dough rheological parameters; work to extend, compression force, and relaxation force at 25 s. The flour IPP was also included in the forward and stepwise selection methods. Thus it seems the flour gluten fraction composition is particularly important to tortilla quality through its impact on dough rheology. It is instructive to note that the classical wheat quality parameters used for bread quality prediction (grain hardness, dough mixing properties, protein content) did not make it to the best model. This indicates a need to develop different predictive quality tests for tortillas and flatbreads. 16
ACCEPTED MANUSCRIPT
3.3. Classification summary
PT
Based on stepwise selection, a total of 127 lines of the 185 lines were successfully used
RI
to validate the model. These included 54 lines that had superior tortilla making properties and 73
SC
lines that produced poor quality tortillas. Eight of the good wheat lines were misclassified as poor performing lines whereas thirteen of the poor lines were misclassified as good. A closer
NU
look at the misclassified lines revealed that six of eight good lines that were misclassified as poor
MA
performing lines that were from the TIA population. This may be partly attributed to the unusual protein fraction composition and rheological properties of some of these TIA lines (Jondiko, et
D
al., 2012). For example, several of these lines possess deletions of various HMW-GS alleles at
TE
the Glu A1, Glu B1 and Glu D1 genome which gives then low IPP and makes their dough highly
AC CE P
extensible (see Table 1 summary). However, other parameters like degree of starch damage and flour particle size distribution that were not investigated in this study but were previously reported to impact tortilla quality (Mao & Flores, 2001; Wang & Flores, 2000), may have also contributed, and should be considered as factors in future model refinement. All said, the TIA population had the largest proportion of lines that produced good quality tortillas (47% versus 36 and 37% for the other populations) (Table 6). This is because the TIA samples are under development for flatbreads market, and have been actively improved (genetically) for protein functionality aimed at addressing the unique requirements for these products. The lower relative suitability of the TXE/UVT and TAM1112 populations was related to their tendency to produce tortillas with smaller diameters (Table 1). This is because these
17
ACCEPTED MANUSCRIPT samples, derived from bread wheat populations, tend to form strong gluten which tends to shrink back after pressing (Waniska, et al., 2004).
PT
Multivariate predictive power of kernel, flour and dough properties provides a potential
RI
approach to determine the end use functionality of early to advance generation wheat varieties
SC
for tortillas. It also provides valuable insight on the key grain composition and dough functionality parameters relevant to tortilla quality, which have not been fully identified to date,
NU
owing to the complexity of predicting negatively correlated tortilla quality parameters (diameter
MA
and flexibility during storage). It is apparent from this work that the gluten protein fraction composition (especially IPP and glutenin:gliadin ratio) in the wheat and how they affect dough
D
rheological parameters are among the most important predictors of tortilla-making potential of
TE
wheat. Thus genetic improvement of wheat for tortillas requires looking at quality parameters not traditionally associated with bread and soft wheat products. Interestingly, the major traditional
AC CE P
wheat quality prediction tools, dough mixing properties, protein content, etc, were not as important to predicting tortilla processing quality. With growing specialty product markets, this work provides a foundation for alternative approaches to establishing wheat quality for nontraditional applications.
Acknowledgement We thank Texas Wheat Producers Board, and Texas A&M AgriLife Research for financial support.
18
ACCEPTED MANUSCRIPT LITERATURE CITED AACC-International. (2010). Approved Methods of Analysis, 11th Ed. In Methods 26-50.01, 39-11.01, 54-40.02 St. Paul, MN, U.S.A.: AACC International.
PT
Alviola, J. N., & Awika, J. M. (2010). Relationship between objective and subjective wheat flour tortilla quality evaluation methods. Cereal Chemistry, 87(5), 481-485.
SC
RI
Alviola, J. N., Jondiko, T., & Awika, J. M. (2010). Effect of cross-linked resistant starch on wheat tortilla quality. Cereal Chemistry, 87(3), 221-225.
NU
Alviola, J. N., Jondiko, T. O., & Awika, J. M. (2012). Effect of strong gluten flour on quality of wheat tortillas fortified with cross-linked resistant starch. Journal of Food Processing and Preservation, 36(1), 38-45.
MA
Alviola, J. N., & Waniska, R. D. (2008). Determining the role of starch in flour tortilla staling using alpha-amylase. Cereal Chemistry, 85(3), 391-396. Barros, F., Alviola, J. N., Tilley, M., Chen, Y. R., Pierucci, V. R. M., & Rooney, L. W. (2010). Predicting hot-press wheat tortilla quality using flour, dough and gluten properties. Journal of Cereal Science, 52(2), 288-294.
TE
D
Bean, S. R., Lyne, R. K., Tilley, K. A., Chung, O. K., & Lookhart, G. L. (1998). A rapid method for quantitation of insoluble polymeric proteins in flour. Cereal Chemistry, 75(3), 374-379.
AC CE P
Bejosano, F. P., Joseph, S., Lopez, R. M., Kelekci, N. N., & Waniska, R. D. (2005). Rheological and sensory evaluation of wheat flour tortillas during storage. Cereal Chemistry, 82(3), 256-263. Bello, A. B., Serna- Saldivar, S. O., Waniska, R. D., & Rooney, L. W. (1991). Methods to prepare and evaluate wheat tortillas. Cereal Foods World, 36, 315-322 Cinco-Moroyoqui, F. J., & MacRitchie, F. (2008). Quantitation of LMW-GS to HMW-GS ratio in wheat flours. Cereal Chemistry, 85(6), 824-829. Friend, C. P., Ross, R. G., Waniska, R. D., & Rooney, L. W. (1995 ). Effects of additives in wheat flour tortillas. Cereal Foods World, 40 494-497. Fu, B. X., & Kovacs, M. I. P. (1999). Rapid single-step procedure for isolating total glutenin proteins of wheat flour. Journal of Cereal Science, 29(2), 113-116. Gupta, R. B., Khan, K., & Macritchie, F. (1993). Biochemical basis of flour properties in bread wheats. I. Effects of variation in the quantity and size distribution of polymeric protein. Journal of Cereal Science, 18(1), 23-41. Jondiko, T. O., Alviola, N. J., Hays, D. B., Ibrahim, A., Tilley, M., & Awika, J. M. (2012). Effect of high molecular weight glutenin subunit allelic composition on wheat flour tortilla quality. Cereal Chemistry, 89(3), 155-161. Mao, Y., & Flores, R. A. (2001). Mechanical Starch Damage Effects on Wheat Flour Tortilla Texture. Cereal Chemistry, 78(3), 286-293. 19
ACCEPTED MANUSCRIPT Mondal, S., Tilley, M., Alviola, J. N., Waniska, R. D., Bean, S. R., Glover, K. D., & Hays, D. B. (2008). Use of near-isogenic wheat lines to determine the glutenin composition and functionality requirements for flour tortillas. Journal of Agricultural and Food Chemistry, 56(1), 179-184.
PT
Pascut, S., Kelekci, N., & Waniska, R. D. (2004). Effects of wheat protein fractions on flour tortilla quality. Cereal Chemistry, 81(1), 38-43.
RI
Pierucci, V. R. M., Tilley, M., Graybosch, R. A., Blechl, A. E., Bean, S. R., & Tilley, K. A. (2009). Effects of overexpression of high molecular weight glutenin subunit 1Dy 10 on wheat tortilla properties. Journal of Agricultural and Food Chemistry, 57(14), 6318-6326.
SC
Smewing, J. (1995). Measurement of dough and gluten extensibility using the SMS/Kieffer Rig and the TA.XT2 Texture Analyzer. Stable Micro Systems Ltd, Godalming(Surrey, UK).
NU
Suchy, J., Lukow, O. M., & Fu, B. X. (2003). Quantification of monomeric and polymeric wheat proteins and the relationship of protein fractions to wheat quality. Journal of the Science of Food and Agriculture, 83(10), 1083-1090.
MA
TIA. (2013). Tortilla Industry Association, Tortilla Industry Overview. In (Vol. 2013): Annual Convention & trade Exposition Technical Conference. http://www.tortilla-info.com/ (Accessed October 2014).
TE
D
Tuncil, Y. E., Jondiko, T., Tilley, M., Hays, D. B., & Awika, J. M. (2016). Combination of null alleles with 7 + 9 allelic pair at Glu-B1 locus on the long arm of group 1 chromosome improves wheat dough functionality for tortillas. LWT - Food Science and Technology, 65, 683-688.
AC CE P
Wang, L., & Flores, R. A. (2000). Effects of Flour Particle Size on the Textural Properties of Flour Tortillas. Journal of Cereal Science, 31(3), 263-272. Waniska, R. D., Cepeda, M., King, B. S., Adams, J. L., Rooney, L. W., Torres, P. I., Lookhart, G. L., Bean, S. R., Wilson, J. D., & Bechtel, D. B. (2004). Effects of flour properties on tortilla qualities. Cereal Foods World, 49(4), 237-244. Zhang, P., Jondiko, T. O., Tilley, M., & Awika, J. M. (2014). Effect of high molecular weight glutenin subunit composition in common wheat on dough properties and steamed bread quality. Journal of the Science of Food and Agriculture, 94(13), 2801-2806.
20
ACCEPTED MANUSCRIPT Table 1 Means, minimum and maximum values for variables measured on the TIA, TXE/UVT and TAM1112 wheat lines1 TAM1112 Min Mean Max
66.7 2.6 28.1
81.1 3.1 38.1
69.4 2.4 26.1
49.4 2.3 21.0
59.6 2.4 25.0
74.3 2.6 28.9
71.0 14.9 61.9 2.9 0.6 0.4 41.4
76.0 16.4 64.1 5.8 0.8 0.6 51.3
0.3 82.9 17.5 1.6 8.4 6.0 104.1
152.0 169.4 1.3 1.7 80.2 82.5 1.3 3.4 0.5 0.8 3.8 6.1 9.6 11.4 13.3 24.6 45
RI
TE
D
0.1 23.1 4.1 1.2 5.7 3.2 73.8
80.3 2.6 30.8
89.9 2.8 34.8
61.6 11.9 58.0 1.5 0.7 0.3 35.5
68.1 13.0 60.6 3.2 0.8 0.3 43.9
75.3 14.4 62.3 6.5 1.0 0.4 53.6
61.3 13.8 61.0 2.0 0.5 0.3 12.2
69.8 14.6 64.6 3.8 0.7 0.3 49.7
77.9 15.4 70.3 7.5 0.8 0.4 72.2
0.8 136.2 31.1 2.7 11.4 11.1 145.3
0.2 37.5 10.8 1.5 5.3 3.2 44.4
0.4 71.1 18.1 1.8 7.8 5.2 89.4
0.8 112.0 24.5 2.0 10.2 7.3 143.2
0.3 24.1 12.7 1.3 6.1 3.8 35.8
0.6 55.9 20.9 1.7 8.7 5.9 91.0
1.2 105.1 30.3 2.7 12.4 8.6 170.3
184.0 2.1 84.8 5.0 1.4 9.8 14.6 47.9
146.3 0.8 72.9 2.1 0.6 5.3 8.3 26.9
164.9 182.7 136.1 163.0 181.1 1.5 1.9 1.1 1.6 2.0 82.2 84.2 72.6 81.9 85.6 3.2 4.3 1.0 3.0 4.5 0.9 1.1 0.5 0.9 1.4 8.5 17.5 4.5 10.7 17.5 16.4 19.6 14.1 19.0 24.7 50.6 87.5 28.1 82.1 202.5 41 101
SC
59.1 13.3 60.3 1.5 0.5 0.3 29.2
AC CE P
Kernel properties Hardness (Index) Diameter (mm) Weight (g) Flour properties Milling yield (%) Protein content (14% mb) 3 Mixo water absorption (%) 4 Mixo time (min) Glutenin to Gliadin Ratio 5 H_L_GS_Ratio 6 IPP (%) Dough rheology * Elasticity (N) Extensibility (mm) Work to extend (N.mm) Relaxation time (min) 7 F_25 (N) 8 F_100 (N) Compression force (N) Tortilla properties Tortilla diameter Tortilla specific volume (cm3/g) Lightness (L) Flexibility Score Gradient (N/mm) Force (N) Distance (mm) Work (N.mm) N
NU
38.2 2.5 24.1
Min
2
MA
TIA Mean
PT
Max
Population TXE_UVT 2012 Min Mean Max
Variables
1
Total number of lines evaluated from each of the three wheat populations including outliers. Minimum and maximum values from each population. 3 Mixograph water absorption values; tortilla water absorption values are 10 points less. 4 Dough development time is the mixograph peak time. 5 High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio. 6 Insoluble Polymeric Protein content (as a proportion of total proteins) 7 Stress Relaxation Force after 25 Sec. and after 8100 Sec. compression. *Elasticity means the force recorded as resistance to extension in dough extensibility test. 2
21
ACCEPTED MANUSCRIPT Table 2
Standard
Intercept
1
-1.584
8.32
Hardness
1
-0.050
0.03
3.48
Diameter
1
9.734
4.50
Weight
1
-0.496
0.21
Protein
1
-1.137
1
1
7.274
2
1
-9.273
3
1
4 5 6 7
1
IPP ForceD Area F_25
0.06
0.95
4.68
0.03
16878.63
5.82
0.02
0.61
0.42
7.20
0.01
0.32
3.06
5.64
0.02
1442.90
5.39
2.96
0.09
0.00
-0.053
0.04
2.02
0.16
0.95
1
1.585
1.74
0.83
0.36
4.88
1
-0.171
0.07
6.44
0.01
0.84
1
1.354
0.31
19.19
<.0001
3.87
0.029
0.01
7.08
0.01
1.03
NU
SC
0.21
AC CE P
CompForce
0.85
MA
H_L_GS_Ratio
Exp(Est)
TE
GGRatio
Pr > ChiSq
RI
Estimate
Parameter
D
DF
Wald ChiSquare 0.04
PT
Forward selection procedure analysis of maximum likelihood estimates
1
Glutenin to Gliadin Ratio.
2
High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio.
3
Insoluble Polymeric Protein content
4
Dough elasticity is the force recorded as resistance to extension in dough extensibility test.
5
Work to extend dough strips from the dough extensibility test.
6
Stress Relaxation Force after 25 Seconds of dough compression.
7
Maximum compression force recorded from the dough compression test.
22
ACCEPTED MANUSCRIPT Table 3 Backward elimination procedure analysis of maximum likelihood estimates
Estimate
Standard
Intercept
1
-5.300
7.76
Wald ChiSquare 0.47
Hardness
1
-0.052
0.03
3.90
Diameter
1
9.920
4.39
Weight
1
-0.479
Protein
1
-1.075
1
1
7.846
2
1
-9.976
3
1
4 5
F_25 CompForce
0.01
0.05
0.95
5.10
0.02
20339.65
0.20
5.70
0.02
0.62
0.41
6.76
0.01
0.34
2.92
7.24
0.01
2554.50
5.45
3.35
0.07
0.00
-0.185
0.06
8.58
0.00
0.83
1
1.410
0.28
24.92
<.0001
4.10
1
0.028
0.01
6.71
0.01
1.03
SC
RI
0.49
NU
Area
Exp(Est)
MA
H_L_GS_Ratio
Pr > ChiSq
D
GGRatio
PT
DF
Parameter
Glutenin to Gliadin Ratio.
2
High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio.
3
Work to extend dough strips from the dough extensibility test.
4
Stress Relaxation Force after 25 Seconds of dough compression.
5
Maximum compression force recorded from the dough compression test.
AC CE P
TE
1
23
ACCEPTED MANUSCRIPT Table 4 Stepwise selection procedure analysis of maximum likelihood estimates
Standard Error
Intercept
1
-10.149
3.35
Weight
1
-0.189
0.08
1
1
10.583
2
1
3 4
Wald ChiSquare
PT
Estimate
Exp(Est)
9.18
0.00
0.00
5.99
0.01
0.83
2.39
19.53
<.0001
39446.00
-0.052
0.03
2.30
0.13
0.95
1
-0.219
0.06
11.82
0.00
0.80
1
1.401
0.32
19.39
<.0001
4.06
5
1
0.355
0.24
2.17
0.14
1.43
6
1
0.021
0.01
4.75
0.03
1.02
Area F_25 F_100 Compression force
SC
IPP
MA
GGRatio
RI
Pr > ChiSq
NU
DF
Parameter
Glutenin to Gliadin Ratio.
2
Insoluble Polymeric Protein content
3
Work to extend dough strips from the dough extensibility test.
4
Stress Relaxation Force after 25 Seconds of dough compression.
5
Stress Relaxation Force after 100 Seconds of dough compression.
6
Maximum compression force recorded from the dough compression test.
AC CE P
TE
D
1
24
ACCEPTED MANUSCRIPT Table 5 Classification summary for cross-validation using five nearest neighbors
POOR
Total Priors
GOOD
POOR
20
61
42
19
67.2
32.8
100
68.9
31.2
25
63
88
19
28.4
71.6
100
21.6
66
83
149
61
44.3
55.7
100
0.41
0.59
0.30
Error
Overall Apparent Error rates 0.28
Error
GOOD
POOR
Total Error
61
46
8
54
100
85.2
14.8
100
69
88
13
60
73
78.4
100
17.8
82.2
100
88
149
59
68
127
40.9
59.1
100
46.5
53.5
100
0.41
0.59
0.43
0.57
0.31
0.22
0.15
0.18
0.26
0.17
TE
D
0.33
Total
AC CE P
Rate
PT
41
Total
SC
GOOD
POOR
NU
Type
GOOD
Stepwise selection
MA
From
Backward elimination
RI
Forward selection
25
ACCEPTED MANUSCRIPT Table 6 Actual number of lines and proportions based on all raw data including outliers 3
Good
Poor
Total
21 (47%)
24 (53%)
TXE/UVT
15 (37%)
26 (63%)
TAM1112
36 (36%)
Total
72 (39%)
45
SC
RI
TIA
41
101
115 (61%)
187
NU
64 (64%)
TIA, specialty lines under development for flatbreads; TXE/UVT, advanced lines and varieties
MA
1
2
Population
PT
1
under development as bread wheat; TAM1112, experimental lines from TAM 111 and TAM 112
Good and Poor means number of lines that were suitable, and not suitable for production of
AC CE P
tortillas based on actual data.
TE
2, 3
D
populations.
26
ACCEPTED MANUSCRIPT
PT
Industrial Relevance
RI
Tortillas and other flatbread manufacturers currently use wheat developed for other commodities and rely
SC
on trial and error, and use of various additives to optimize product quality. Genetic development of wheat for these markets is impeded by lack of knowledge of specific grain quality parameters to target. With the
NU
growing demand for clean label and healthy offerings by consumers, the industry is looking for natural ingredients with improved functionality. This work provides the first insight into the specific wheat
AC CE P
TE
D
MA
composition and functional parameters that can predict tortilla quality.
27
ACCEPTED MANUSCRIPT
PT
Highlights
Traditional bread wheat quality tests poor predictors of functionality in tortilla.
-
Multivariate approach used to identify key wheat quality parameters for tortillas.
-
Glutenin-giladin ratio and insoluble polymeric protein major compositional predictors.
-
Dough extensibility and stress relaxation important functional predictors.
-
Stepwise regression model with 83% prediction efficiency developed.
AC CE P
TE
D
MA
NU
SC
RI
-
28
S1466-8564(16)00016-3 doi: 10.1016/j.ifset.2016.01.010 INNFOO 1464
To appear in:
Innovative Food Science and Emerging Technologies
Received date: Revised date: Accepted date:
12 October 2015 31 December 2015 1 January 2016
Please cite this article as: Jondiko, T.O., Yang, L., Hays, D.B., Ibrahim, A.M.H., Tilley, M. & Awika, J.M., Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties, Innovative Food Science and Emerging Technologies (2016), doi: 10.1016/j.ifset.2016.01.010
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT Prediction of wheat tortilla quality using multivariate modeling of kernel, flour, and dough properties
RI
PT
Tom O. Jondikoa,b, Liyi Yanga,b, Dirk B. Haysa, Amir M.H. Ibrahima, Michael Tilleyc, and Joseph M. Awikaa,b,*
Department of Soil and Crop Sciences, Texas A&M University, College Station, TX, 77843
b
Department of Nutrition and Food Science, Texas A&M University, College Station, TX, 77843
c
USDA-ARS Center for Grain and Animal Heatlh Research, 1515 College Avenue, Manhattan, KS 66502, USA
MA
NU
SC
a
AC CE P
TE
D
*Corresponding author: Email: [email protected]; Phone: 979-845-2985, Fax: +1 979-845-0456.
The use of trade, firm, or corporation names in this presentation is for the information and convenience of the reader. Such use does not constitute an official endorsement or approval by the United States Department of Agriculture or the Agricultural Research Service of any product or service to the exclusion of others that may be suitable.
1
ACCEPTED MANUSCRIPT Abstract
Traditional wheat quality methods for bread have poor predictive power for flatbreads quality,
PT
which impedes genetic improvement of wheat for the growing market. We used a multivariate discriminant analysis to predict tortilla quality using a set of 16 variables derived from kernel
RI
properties, flour composition, and dough rheological properties of 187 experimental hard wheat
SC
samples grown across Texas. A discriminant rule (suitability for tortillas = diameter > 165mm +
NU
day 16 flexibility score >3.0) was used to classify samples. Multivariate normal distribution of the data was established (Shapiro-Wilk p > 0.05). Logistic regression and stepwise variable
MA
selection identified an optimum model comprising kernel weight, glutenin-gliadin ratio, insoluble polymeric proteins, and dough extensibility and stress relaxation parameters, as the
D
most important variables. Cross-validation indicated 83% prediction efficiency for the model.
TE
This work provides important insight on potential targets for wheat quality genetic improvement
AC CE P
for tortillas and specialty products market.
Keywords Wheat quality; tortilla; flatbread; gluten; genetic improvement
2
ACCEPTED MANUSCRIPT 1. Introduction Tortillas and other specialty breads are increasingly becoming a global dietary staple. For
PT
example, according to the Tortilla Industry Association report for 2013, wheat tortilla was the
RI
only bakery segment that experienced growth in 2012 and is projected to increase further (TIA, 2013). Wheat tortilla sales exceeded $ 6 billion in 2012, affirming consumer preference for its
SC
versatility and functional convenience. To consumers, the definition of good quality tortilla
NU
encompasses its ability to retain flexibility and be large enough to wrap food (Waniska, et al., 2004).
MA
Despite the growing popularity of tortillas, the main challenge is that there is no reliable and practical method to predict wheat (Triticum aestivum L.) functionality for tortillas, as is the
TE
D
case with pan bread and other mainstream baked products. Currently, most tortilla ingredient suppliers and processors use trial and error and additives to optimize tortilla quality, which
AC CE P
compromise sensory appeal and quality, and adds to cost of manufacture. The health conscious consumer is also increasingly demanding use of fewer additives (clean label), as well as healthier options, like whole grains. Thus there is a need to develop wheats with improved functionality for the tortilla and flatbreads market. However, effective genetic improvement efforts require that the key grain quality parameters responsible for desirable product characteristics can be reliably identified. Ideally the methods to identify these grain quality factors should be rapid and require small sample size. Currently, the only way to predict wheat functionality for tortillas is to actually make the product. Tortilla quality in the US is largely defined by diameter, flexibility during storage, and opacity (Alviola & Awika, 2010; Alviola, Jondiko, & Awika, 2010). Earlier studies reported that tortillas of good quality can be produced using wheat flour of intermediate protein content, 3
ACCEPTED MANUSCRIPT protein strength and low level of starch damage by (Waniska, et al., 2004). However, the vast majority of wheat that fall in this category did not produce a good quality product (Waniska, et
PT
al., 2004). Tortilla diameter can be predicted using linear equations comprising mixing time and
RI
dough resistance to extension (Barros, et al., 2010). However, these linear models did not significantly predict tortilla flexibility, which is a critical tortilla quality attribute. A major
SC
challenge to predicting tortilla quality is the negative correlation that generally exists between
NU
the main quality factors, tortilla diameter, and flexibility during storage (Mondal, et al., 2008; Pascut, Kelekci, & Waniska, 2004). Large diameter generally requires weak and extensible
MA
dough, which tends to be detrimental to tortilla flexibility. Genetic improvement of wheat targeting alterations of high molecular weight glutenin
TE
D
sub-units (HMW-GS) composition through deletion at one or more of the Glu A1, B1, and D1 loci has shown promise as a way to produce unique gluten functionality ideal for tortillas and
AC CE P
other specialty products (Jondiko, et al., 2012; Mondal, et al., 2008; Tuncil, et al., 2016; Zhang, et al., 2014). There is a need to develop prediction models that can be used to screen these wheat varieties under development for end use functionality. Hence, this study used multivariate statistical methods designed to elicit information from simultaneous measurements of many variables acquired from wheat kernel, flour, and dough to predict the quality of tortilla, specifically the diameter and flexibility during storage. In addition, the study investigated whether these multivariate models could be used to reliably classify early to late generation wheat lines for their potential to produce good quality tortillas. The objectives of this study were to develop a model that can be used to predict the functional performance of wheat varieties for tortilla processing, and provide insight on wheat quality parameters that can be targeted for genetic improvement for specialty flatbreads market. 4
ACCEPTED MANUSCRIPT
2. Materials and Methods
PT
2.1. Experimental materials
RI
A total of 187 hard winter wheat (HWW) lines were used. These included advanced generation breeding lines comprising 38 Texas elite (TXE) lines, 3 uniform variety trial (UVT)
SC
lines; 101 experimental lines derived from ‘TAM111’/‘TAM112’ population (TAM1112); and
NU
45 identity preserved lines specifically developed for specialty flat breads (TIA). These TIA lines were selected based on variations in their allelic composition at the HMW-GS loci Glu A1, Glu
TE
2.2. Kernel properties and milling
D
across Texas in 2009 – 2012 seasons.
MA
B1 and Glu D1. The wheat samples were grown in Texas A&M AgriLife experimental plots
The wheat lines were evaluated for hardness, diameter, weight and moisture content
AC CE P
using a single kernel hardness tester (SKCS) (Perten Instruments, Springfield, IL). The grains were tempered to 14% moisture content (AACC Method 26-50.01) and milled using a Quadrumat Senior mill (Brabender Instruments, South Hackensack, NJ). Flour milling yields were recorded. Samples were processed into tortillas with 1- 2 months after milling. 2.3. Evaluation of flour properties 2.3.1. Total protein content. Total flour protein content was determined in two replicates for each wheat line using near-infrared reflectance spectroscopy (Perten PDA 7000 Dual Array with Grams Software) according to AACC Method 39-11.01 (AACC-International, 2010).
2.3.2. Polymeric to monomeric protein ratio (glutenin to gliadin ratio). Protein extraction of proteins followed the method of Gupta, Khan, & Macritchie (1993). Briefly, a 10 mg flour sample were mixed with 1 mL 0.05 M Sodium phosphate buffer, pH 6.9, containing 0.5% SDS 5
ACCEPTED MANUSCRIPT (w/v) then sonicated for 35 s at power setting 10 W. The sample was then be centrifuged at 15,000 × g for 5 min and the supernatant collected (contains total protein) and filtered through
PT
0.45 μm filter and analyzed by size – exclusion HPLC using an Agilent 1260 HPLC (Agilent,
RI
Santa Clara, CA), 300 x 7.8 mm BioSep-SEC-S400 column (Phenomenex, Torrance, CA) with
gradient system composed of 50% ACN+0.1% TFA (B) and 50% water+ 0.1% TFA (A), 30°C
SC
column temp., at a flow rate of 1 ml/min for 30 min run. The chromatograms were manually
NU
integrated. The area of the first peak corresponds to total polymeric proteins and the area of the second peak to monomeric proteins (Gupta, Khan, & Macritchie, 1993). Two replicates of each
MA
flour sample were analyzed.
D
2.3.3. Insoluble polymeric protein content (IPP). Ten mg of flour was suspended in 1 mL of 0.05
TE
M sodium phosphate buffer (pH 6.9), containing 0.5% sodium dodecyl sulfate (SDS) and shaken on a vortex for 30 min. The mixture then centrifuged for 5 min at 16,000 × g. The supernatant
AC CE P
(containing soluble polymeric protein - SPP) were collected and filtered (0.45 μm) and analyzed by size – exclusion HPLC as described above (Bean, et al., 1998). The pellet were mixed with 1 ml sodium phosphate buffer and sonicated for 25 s at 10 W. The mixture were centrifuged at 16,000 × g /5 min, and the supernatant collected and filtered as above then analyzed using the SE-HPLC as described above. The percentages of soluble (extractable) and insoluble (unextractable) polymeric protein were calculated as [peak 1 area (extractable)/peak 1 area (total)] × 100 and [peak 1 area (unextractable)/peak 1 area (total)] × 100 respectively. Peak 1 (total) refers to the sum of peak 1 (extractable) and peak 1 (unextractable). 2.3.4. High molecular weight to low molecular weight glutenin sub-units ratio (H_L_GS_Ratio). HMW-GS and LMW-GS were quantified using RP-HPLC. A sample of 100 mg flour was mixed with 1 mL sodium iodate buffer (0.3 M sodium iodate + 7.5% isopropanol) and vortexed for 15 6
ACCEPTED MANUSCRIPT min. The mixture was centrifuged for 5 min at 15,000 × g. The supernatant containing gliadins were discarded. To the pellet 1ml water were added then shaken for 5 min and centrifuged at
PT
15,000 x g for 5 min. The pellet were mixed with 1 mL 50% isopropanol containing 2% BME and vortex for 30 min, and then centrifuged for 5 min. at 15,000 × g. The supernatant was
SC
RI
collected (contains glutenins) and 600 L of the glutenin extract was alkylated with 40 µL 4vinylpyridine for 15 min at 60°C. The resulting sample was injected into a Phenomenex Jupiter
NU
C18 250 × 4.6 mm diameter column, 5 μ particle size, and 300 Å pore size. The solvent flow rate was 1.0 mL/min and composed of water (A) and acetonitrile (B), both containing 0.1% TFA.
MA
The gradient was as follows: 0-3 min from 25% B to 35% B, 3-24 min increased to 53%B, the gradient decreased to 25% B at 25 min and kept at 25% B until 29 min. Detection of protein
TE
D
peaks were carried out by UV detector at 210 nm. The area of the curve corresponding to HMWGS and LMW-GS contents were determined by manual integration and the HMW/ LMW-GS
Fu, 2003).
AC CE P
ratio calculated (Cinco-Moroyoqui & MacRitchie, 2008; Fu & Kovacs, 1999; Suchy, Lukow, &
2.4. Dough mixing and rheology
2.4.1. Dough development time. A Mixograph (National Manufacturing Co., Lincoln, NE) was used to estimate dough mixing/ development time and flour water absorption; ten grams of flour were used (14% mb) (AACCI Method 54-40.02; AACCI 2010). Mixing/ dough development time was manually calculated from the mixograph by drawing two midlines from each end of the graph. The point of crossover was marked as the peak time for each wheat line (Alviola & Awika, 2010).
7
ACCEPTED MANUSCRIPT 2.4.2. Dough compression force. Dough compression test was used to measure the maximum compression force required to deform a 45 g dough ball. Two dough balls were compressed to
PT
70% of their height using 10 cm diameter probe on a TA-XT2 texture analyzer (Texture
RI
Technologies Corp., Scarsdale, NY). Maximum dough compression force was calculated as an
SC
average force of the two dough balls for each of the wheat lines. (Bejosano, et al., 2005)
NU
2.4.3. Stress Relaxation test. Stress relaxation tests were conducted by compressing two 45 g
MA
dough balls on a TA-XT2 texture Analyzer. This was conducted on dough balls that had been rested for10 min. A 10 cm diameter cylindrical probe was
lowered at 1 mm/s on each dough
D
ball and held for 120 s. The relaxation force at 25 s, 100 s, maximum force, and relaxation time
TE
were recorded using the texture expert software (Jondiko, et al., 2012).
AC CE P
2.4.4. Dough extensibility evaluation. Dough extensibility properties: elasticity (N), extensibility (mm) and work to extend (N.mm) were determined using the modified dough extensibility test (Jondiko, et al., 2012). Doughs were prepared using 100 g flour and 2 g salt for each of the wheat lines and mixed in a Mixograph 100 g micro-mixer. Warm water (~ 35º C) was added to the dry ingredients based on an adjusted mixograph water absorption value and mixed to optimum dough development time, based on the mixograph test (Jondiko, et al., 2012). Dough extensibility parameters were evaluated on 10 dough strips from each wheat line as previously described (Smewing, 1995).
2.5. Tortilla formulation and processing
8
ACCEPTED MANUSCRIPT Tortilla formulation included two batches of 500 g flour from each of the wheat lines, without reducing agents as described (Alviola, Jondiko, & Awika, 2012). The quantity of
PT
distilled water used was based on mixograph water absorption for each line, minus 10% points.
RI
Mixed dough was divided and rounded into approx. 45 g dough balls (Duchess Divider/Rounder, Bakery Equipment and Service Co., San Antonio, TX). The dough balls were rested at 32º C and
SC
65-70% relative humidity in the proofing chamber for 10 minutes, and then hot-pressed (205°C,
NU
7.9 kPa, 1.4 sec) into round discs (Bello, et al., 1991) and baked for 30 sec at 190 and 200 °C in a gas fired three-tier oven (Model 0P01004-02, Lawrence Equipment, El Monte, CA). Tortillas
MA
were then cooled to room temp and packed in 1 mm thick polyethylene bags and stored at 22oC
TE
D
for textural and shelf stability evaluation.
AC CE P
2.6. Tortilla quality evaluation
2.6.1. Tortilla physical property evaluation. Tortilla physical properties: weight, diameter, and thickness were evaluated on ten randomly selected tortillas from each of the wheat lines following the method of Friend and others (Friend, et al., 1995 ). These properties were used to determine the tortilla specific volume as described below: ). Where; height = height of a single tortilla (cm); weight = weight of a single tortilla (g), r = average radius of a tortilla (cm) (Waniska, et al., 2004). Tortilla lightness (L* – values) was measured on two randomly selected tortillas at two different points of each side of the tortillas using a Minolta color meter (Chroma Meter CR-310, Minolta, Tokyo, Japan). 9
ACCEPTED MANUSCRIPT 2.6.2. Subjective tortilla shelf stability evaluation. Tortilla rollability/ flexibility score was determined after 16 days of storage, using procedures recently described (Alviola & Waniska,
RI
indicative of inferior/ poor shelf stability (Alviola, et al., 2008).
PT
2008; Jondiko, et al., 2012). Flexibility scores < 3 (numerous cracks and breaks on tortillas) was
SC
2.6.3. Objective tortilla shelf stability evaluation. The two-dimensional extensibility test (Bejosano, et al., 2005) was used to measure tortilla textural changes after day 0, 4, 8, 12 and 16
NU
of storage. The extensibility test was determined on four tortillas from each of the wheat lines at
MA
each time point using a TA-XT2i texture analyzer. The analyzer settings were: return to start option, trigger force (0.05 N), pre and post-test speeds (10.0 mm/s) and test speed (1.0 mm/s).
D
The texture expert software was used to record tortilla texture properties: Gradient/ modulus of
AC CE P
TE
deformation (N/mm), force (N), distance (mm) and work to rupture (N.mm).
2.7. Statistical analysis
2.7.1. Data description and classification criterion. A total of 16 continuous variables were measured on the 187 wheat lines using the methods described above. The 45 wheat lines possessing variations in high molecular weight glutenin sub-units (HMW-GS) at the Glu A1, Glu B1 and Glu D1 were used to train and calibrate the prediction models. This was based on the evidence that these lines had the unique protein functionality to produce tortillas of superior quality (Jondiko, et al., 2012). Sixteen variables measured on all the 187 lines were then used to cross validate the prediction models. These variables were kernel properties (hardness, diameter, and weight); flour properties (protein content, mixograph water absorption, glutenin to gliadin ratio (GGRatio), high molecular weight to low molecular weight glutenin sub-unit ratio 10
ACCEPTED MANUSCRIPT (H_L_GS_Ratio), insoluble polymeric protein content (IPP)); and dough rheology (dough development/ peak time (Mixo_Time), dough compression force (CompForce), resistance to
PT
extension (ForceD), extensibility (DistanceD), work to extend (WORK), stress relaxation time
RI
(RT), stress relaxation force after 25 s (F_25) and after 100 s (F100) of compression). Predication models were developed for both tortilla diameter and day 16 flexibility scores using
SC
the 16 variables measured on wheat kernel, flour and dough rheology.
NU
A suitability of categorical numbers 1 or 0 was assigned to each wheat line, where varieties with “1” were suitable to make quality tortillas (Diameter ≥ 165 cm and day 16
MA
flexibility score ≥ 3.0). Wheat lines not meeting this description were assigned “0” meaning they
D
were not suitable to make quality tortillas.
TE
2.7.2. Test for data normality. Multivariate normal distribution of the data set was determined
AC CE P
using the Shapiro-Wilk statistic p – value for each of the 23 variables (p < 0.05). 2.7.3. Discriminant analysis using logistic regression. Discriminant analysis was done using SAS version 9.2 (SAS Institute, Cary, NC). The data was standardized to eliminate the effect of differences in units of each variable. In order to predict wheat lines suitable to make good quality tortillas, a discriminant analysis were carried out using all the wheat kernel, flour, and dough property variables. Prediction models were developed based on three methods of variable selection. These were forward, backward, and stepwise variable selection. Each model was used to classify all the wheat lines into two suitability classes based on a discriminant rule where if the response (Suitability) based on the variables in each model was equal to one (= 1) for a line that was good for tortilla processing , otherwise the wheat line had inferior tortilla processing ability (response was 0). The models were used to cross validate prediction accuracy using five nearest 11
ACCEPTED MANUSCRIPT neighbors in logistic regression discriminant analysis. An apparent error rate of classification was calculated by dividing the number of lines that were misclassified with the total number of
RI
PT
lines evaluated during cross validation.
SC
3. Results and discussion
NU
3.1. Data distribution and normality
The means, minimum and maximum values for kernel, flour, dough and tortilla
MA
properties for the wheat lines are given in Table 1. Two of the 187 wheat lines were identified as
D
outliers using principal component analysis and eliminated hence, the total number of
TE
observations were N = 185 for all the 16 variables (Table 1). To eliminate the effect of differences in units in the variables, data analysis was conducted on standardized variables. The
AC CE P
variables were standardized using a fixed mean and variance. Kernel property values had a broad range (Table 1) and followed a multivariate normal distribution (p < 0.05) indicating good diversity in the wheat lines. The mean hardness for the TIA, TXE/UVT and TAM1112 lines ranged between 38 and 89. Lines from the TXE/UVT populations had the highest average hardness index (80.3, vs 66.7 for TIA and 59.6 for TAM1112); this was expected because these are advanced materials that have gone through extensive selection towards potential release as hard wheat varieties. Flour protein content (as is) ranged between 11.9 and 16.4%, whereas mixograph mixing time ranged between 1.5 and 7.5 min, with no obvious differences among the different wheat populations. These ranges were broad enough to capture expected variability in wheat quality. Other traits of interest, like IPP content and dough extensibility, that have been shown to influence tortilla quality (Jondiko, et 12
ACCEPTED MANUSCRIPT al., 2012), also had wide ranges (Table 1). Notably, dough extensibility, which correlates positively with tortilla diameter (Jondiko, et al., 2012; Pierucci, et al., 2009) was on average
PT
highest for the TIA lines (82.9 mm, vs 71.1 for TXE/UVT and 55.9 mm for TAM 1112). The
RI
high extensibility of these TIA lines has been previously demonstrated, and is related to the alteration in their HMGS alleles through specific deletions at Glu A1, B1 or D1 genome
SC
(Jondiko, et al., 2012).
NU
Tortilla quality attributes also had a broad range of distribution, representing very poor to
MA
excellent product quality. For example, diameter values ranged from 136 – 184 mm, and day 16 flexibility scores ranged from 1.0 (poor) to 5.0 (excellent) (Table 1). A diameter value of 165
D
mm and a day 16 flexibility score of 3.0 are the minimum values for defining good quality
TE
tortillas based on the protocols used in this study (Bejosano, et al., 2005). Tortilla specific volume (an indicator of leavening effect) ranged between 0.8 and 2.1 cm3/g, whereas lightness
AC CE P
(L value) ranged from 72.6 – 85.6; the L-value is a good indicator of tortilla opacity, with higher values being preferred by USA consumers (Alviola & Awika, 2010; Alviola, et al., 2012). Among the different wheat populations used, the TIA samples had higher average diameter and day 16 flexibility score than the TXE-UVT and the TAM1112 lines (Table 1), which indicates improved functionality of the TIA lines for tortilla production. Diameter typically negatively correlates with flexibility score for tortillas (Pascut, et al., 2004). All tortilla quality parameters were normally distributed (p > 0.05).
3.2. Discriminant analysis and model selection
13
ACCEPTED MANUSCRIPT Variable selection was carried out using forward, backward and stepwise variable selection resulted into three models (Table 2 – 4). Forward selection procedure resulted in a
PT
model with 11 of the 16 variables (Table 2), a total of 149 wheat lines of the 185 lines were used
RI
in this model because 36 lines were eliminated for being outliers based on the variables selected
SC
for the model (Table 5).
Predicted suitability = -1.584 -0.05*Hardness + 9.734*Diameter -0.496*Weight -
NU
1.137*Protein + 7.274*GGRatio - 9.2731*H_L_GS_Ratio -0.053*IPP +1.5851*ForceD -
MA
0.171*Area + 1.354*F_25 + 0.029*CompForce (Table 2). The three kernel properties (hardness, diameter and weight), four flour protein variables
D
(protein, GGRatio, H_L_GS_Ratio and IPP) and four dough rheology variables (extensibility,
TE
work to extend, relaxation force after 25 sec and compression force) were used to classify and
AC CE P
cross validate the model using five nearest neighbors. The apparent error rate for this model was 0.30 (Table 5). This model correctly classified 67% and 72% of the Good and Poor wheat lines with an overall accuracy of 70%. The model was not ideal because it included a lot of variables that are related. For example, both kernel diameter and hardness contribute to kernel weight, whereas dough resistance to extension (ForceD) is a component of work to extend (Area). Backward elimination procedure resulted in the selection of 9 variables for the prediction model (Table 3). One hundred and forty nine lines were used in validating this model after elimination of outliers based on the variables selected. The backward elimination model was as follows:
14
ACCEPTED MANUSCRIPT Predicted suitability = -5.3 -0.052*Hardness + 9.92*Diameter -0.479*Weight 1.075*Protein + 7.846*GGRatio -9.976*H_L_GS_Ratio -0.185*Area + 1.41*F_25 +
PT
0.028*CompForce (Table 3).
RI
This model had a slightly lower apparent error rate (0.26) compared to the forward
SC
selection model. It misclassified 31 % and 22 % of the good and poor wheat lines respectively. It correctly classified 74 % of the wheat lines (Table 5). The backward elimination procedure
NU
resulted in more efficient selection of variables compared to forward selection procedure; it
MA
resulted in better prediction power with fewer variables. It also eliminated one of the redundant variables, ForceD.
D
Stepwise variable selection model had the fewest variables, seven. A total of 127 lines
AC CE P
5):
TE
were used to validate this model after elimination of outliers based on the seven variables (Table
Predicted suitability = - 10.149 -0.189*Weight + 10.583*GGRatio - -0.052*IPP 0.219*Area + 1.401*F_25 + 0.355*F_100 + 0.021*CompForce (Table 4). This model had one variable each from the kernel properties (Weight), dough extensibility test (Work to extend), and compression test (Compression force) plus two variables from both the flour protein fractions (GGRatio and IPP) and stress relaxation tests (F_25 and F_100). The stepwise model had a significantly lower apparent error rate (0.17) compared to forward and backward elimination models. It correctly classified 83% of the wheat varieties based on their tortilla processing functionality (Table 5). This model was the most efficient of the three, it eliminated most of the correlated variables (grain hardness, and diameter; dough resistance to extension), while providing the best prediction power. 15
ACCEPTED MANUSCRIPT In summary, the forward selection model was the least efficient with an apparent error rate of 0.30. This model requires 11 variables which include several related (thus potentially
PT
redundant) variables. Backward elimination model had nine variables and apparent error rate of
RI
0.26. It misclassified 38 of the wheat varieties, whereas forward selection model misclassified 45. Compared to the forward selection model, it was somewhat more efficient, though it
SC
excluded one of the variables that has been consistently reported to impact tortilla quality
NU
parameters, the IPP, while retaining several correlated variable (kernel hardness, diameter, and weight). On the other hand, it eliminated one of the redundant variables, dough resistance to
MA
extension (ForceD). Thus for practical purposes, its advantage over the forward selection model would be marginal. The stepwise model was the most efficient with an apparent error rate of 0.17
TE
D
(83% prediction accuracy), and the fewest number of variables (seven). It misclassified 21 wheat samples. It also excluded all of the redundant variables that were included in the other two
AC CE P
procedures, while retaining key variables like IPP and work to extend dough. This model provides a good foundation for identifying key wheat quality parameters for tortillas, and predicting the suitability of a wheat line for tortilla production. It is worth noting that five variables showed up in all the three models including, grain weight, flour glutenin-gliadin ration, and three dough rheological parameters; work to extend, compression force, and relaxation force at 25 s. The flour IPP was also included in the forward and stepwise selection methods. Thus it seems the flour gluten fraction composition is particularly important to tortilla quality through its impact on dough rheology. It is instructive to note that the classical wheat quality parameters used for bread quality prediction (grain hardness, dough mixing properties, protein content) did not make it to the best model. This indicates a need to develop different predictive quality tests for tortillas and flatbreads. 16
ACCEPTED MANUSCRIPT
3.3. Classification summary
PT
Based on stepwise selection, a total of 127 lines of the 185 lines were successfully used
RI
to validate the model. These included 54 lines that had superior tortilla making properties and 73
SC
lines that produced poor quality tortillas. Eight of the good wheat lines were misclassified as poor performing lines whereas thirteen of the poor lines were misclassified as good. A closer
NU
look at the misclassified lines revealed that six of eight good lines that were misclassified as poor
MA
performing lines that were from the TIA population. This may be partly attributed to the unusual protein fraction composition and rheological properties of some of these TIA lines (Jondiko, et
D
al., 2012). For example, several of these lines possess deletions of various HMW-GS alleles at
TE
the Glu A1, Glu B1 and Glu D1 genome which gives then low IPP and makes their dough highly
AC CE P
extensible (see Table 1 summary). However, other parameters like degree of starch damage and flour particle size distribution that were not investigated in this study but were previously reported to impact tortilla quality (Mao & Flores, 2001; Wang & Flores, 2000), may have also contributed, and should be considered as factors in future model refinement. All said, the TIA population had the largest proportion of lines that produced good quality tortillas (47% versus 36 and 37% for the other populations) (Table 6). This is because the TIA samples are under development for flatbreads market, and have been actively improved (genetically) for protein functionality aimed at addressing the unique requirements for these products. The lower relative suitability of the TXE/UVT and TAM1112 populations was related to their tendency to produce tortillas with smaller diameters (Table 1). This is because these
17
ACCEPTED MANUSCRIPT samples, derived from bread wheat populations, tend to form strong gluten which tends to shrink back after pressing (Waniska, et al., 2004).
PT
Multivariate predictive power of kernel, flour and dough properties provides a potential
RI
approach to determine the end use functionality of early to advance generation wheat varieties
SC
for tortillas. It also provides valuable insight on the key grain composition and dough functionality parameters relevant to tortilla quality, which have not been fully identified to date,
NU
owing to the complexity of predicting negatively correlated tortilla quality parameters (diameter
MA
and flexibility during storage). It is apparent from this work that the gluten protein fraction composition (especially IPP and glutenin:gliadin ratio) in the wheat and how they affect dough
D
rheological parameters are among the most important predictors of tortilla-making potential of
TE
wheat. Thus genetic improvement of wheat for tortillas requires looking at quality parameters not traditionally associated with bread and soft wheat products. Interestingly, the major traditional
AC CE P
wheat quality prediction tools, dough mixing properties, protein content, etc, were not as important to predicting tortilla processing quality. With growing specialty product markets, this work provides a foundation for alternative approaches to establishing wheat quality for nontraditional applications.
Acknowledgement We thank Texas Wheat Producers Board, and Texas A&M AgriLife Research for financial support.
18
ACCEPTED MANUSCRIPT LITERATURE CITED AACC-International. (2010). Approved Methods of Analysis, 11th Ed. In Methods 26-50.01, 39-11.01, 54-40.02 St. Paul, MN, U.S.A.: AACC International.
PT
Alviola, J. N., & Awika, J. M. (2010). Relationship between objective and subjective wheat flour tortilla quality evaluation methods. Cereal Chemistry, 87(5), 481-485.
SC
RI
Alviola, J. N., Jondiko, T., & Awika, J. M. (2010). Effect of cross-linked resistant starch on wheat tortilla quality. Cereal Chemistry, 87(3), 221-225.
NU
Alviola, J. N., Jondiko, T. O., & Awika, J. M. (2012). Effect of strong gluten flour on quality of wheat tortillas fortified with cross-linked resistant starch. Journal of Food Processing and Preservation, 36(1), 38-45.
MA
Alviola, J. N., & Waniska, R. D. (2008). Determining the role of starch in flour tortilla staling using alpha-amylase. Cereal Chemistry, 85(3), 391-396. Barros, F., Alviola, J. N., Tilley, M., Chen, Y. R., Pierucci, V. R. M., & Rooney, L. W. (2010). Predicting hot-press wheat tortilla quality using flour, dough and gluten properties. Journal of Cereal Science, 52(2), 288-294.
TE
D
Bean, S. R., Lyne, R. K., Tilley, K. A., Chung, O. K., & Lookhart, G. L. (1998). A rapid method for quantitation of insoluble polymeric proteins in flour. Cereal Chemistry, 75(3), 374-379.
AC CE P
Bejosano, F. P., Joseph, S., Lopez, R. M., Kelekci, N. N., & Waniska, R. D. (2005). Rheological and sensory evaluation of wheat flour tortillas during storage. Cereal Chemistry, 82(3), 256-263. Bello, A. B., Serna- Saldivar, S. O., Waniska, R. D., & Rooney, L. W. (1991). Methods to prepare and evaluate wheat tortillas. Cereal Foods World, 36, 315-322 Cinco-Moroyoqui, F. J., & MacRitchie, F. (2008). Quantitation of LMW-GS to HMW-GS ratio in wheat flours. Cereal Chemistry, 85(6), 824-829. Friend, C. P., Ross, R. G., Waniska, R. D., & Rooney, L. W. (1995 ). Effects of additives in wheat flour tortillas. Cereal Foods World, 40 494-497. Fu, B. X., & Kovacs, M. I. P. (1999). Rapid single-step procedure for isolating total glutenin proteins of wheat flour. Journal of Cereal Science, 29(2), 113-116. Gupta, R. B., Khan, K., & Macritchie, F. (1993). Biochemical basis of flour properties in bread wheats. I. Effects of variation in the quantity and size distribution of polymeric protein. Journal of Cereal Science, 18(1), 23-41. Jondiko, T. O., Alviola, N. J., Hays, D. B., Ibrahim, A., Tilley, M., & Awika, J. M. (2012). Effect of high molecular weight glutenin subunit allelic composition on wheat flour tortilla quality. Cereal Chemistry, 89(3), 155-161. Mao, Y., & Flores, R. A. (2001). Mechanical Starch Damage Effects on Wheat Flour Tortilla Texture. Cereal Chemistry, 78(3), 286-293. 19
ACCEPTED MANUSCRIPT Mondal, S., Tilley, M., Alviola, J. N., Waniska, R. D., Bean, S. R., Glover, K. D., & Hays, D. B. (2008). Use of near-isogenic wheat lines to determine the glutenin composition and functionality requirements for flour tortillas. Journal of Agricultural and Food Chemistry, 56(1), 179-184.
PT
Pascut, S., Kelekci, N., & Waniska, R. D. (2004). Effects of wheat protein fractions on flour tortilla quality. Cereal Chemistry, 81(1), 38-43.
RI
Pierucci, V. R. M., Tilley, M., Graybosch, R. A., Blechl, A. E., Bean, S. R., & Tilley, K. A. (2009). Effects of overexpression of high molecular weight glutenin subunit 1Dy 10 on wheat tortilla properties. Journal of Agricultural and Food Chemistry, 57(14), 6318-6326.
SC
Smewing, J. (1995). Measurement of dough and gluten extensibility using the SMS/Kieffer Rig and the TA.XT2 Texture Analyzer. Stable Micro Systems Ltd, Godalming(Surrey, UK).
NU
Suchy, J., Lukow, O. M., & Fu, B. X. (2003). Quantification of monomeric and polymeric wheat proteins and the relationship of protein fractions to wheat quality. Journal of the Science of Food and Agriculture, 83(10), 1083-1090.
MA
TIA. (2013). Tortilla Industry Association, Tortilla Industry Overview. In (Vol. 2013): Annual Convention & trade Exposition Technical Conference. http://www.tortilla-info.com/ (Accessed October 2014).
TE
D
Tuncil, Y. E., Jondiko, T., Tilley, M., Hays, D. B., & Awika, J. M. (2016). Combination of null alleles with 7 + 9 allelic pair at Glu-B1 locus on the long arm of group 1 chromosome improves wheat dough functionality for tortillas. LWT - Food Science and Technology, 65, 683-688.
AC CE P
Wang, L., & Flores, R. A. (2000). Effects of Flour Particle Size on the Textural Properties of Flour Tortillas. Journal of Cereal Science, 31(3), 263-272. Waniska, R. D., Cepeda, M., King, B. S., Adams, J. L., Rooney, L. W., Torres, P. I., Lookhart, G. L., Bean, S. R., Wilson, J. D., & Bechtel, D. B. (2004). Effects of flour properties on tortilla qualities. Cereal Foods World, 49(4), 237-244. Zhang, P., Jondiko, T. O., Tilley, M., & Awika, J. M. (2014). Effect of high molecular weight glutenin subunit composition in common wheat on dough properties and steamed bread quality. Journal of the Science of Food and Agriculture, 94(13), 2801-2806.
20
ACCEPTED MANUSCRIPT Table 1 Means, minimum and maximum values for variables measured on the TIA, TXE/UVT and TAM1112 wheat lines1 TAM1112 Min Mean Max
66.7 2.6 28.1
81.1 3.1 38.1
69.4 2.4 26.1
49.4 2.3 21.0
59.6 2.4 25.0
74.3 2.6 28.9
71.0 14.9 61.9 2.9 0.6 0.4 41.4
76.0 16.4 64.1 5.8 0.8 0.6 51.3
0.3 82.9 17.5 1.6 8.4 6.0 104.1
152.0 169.4 1.3 1.7 80.2 82.5 1.3 3.4 0.5 0.8 3.8 6.1 9.6 11.4 13.3 24.6 45
RI
TE
D
0.1 23.1 4.1 1.2 5.7 3.2 73.8
80.3 2.6 30.8
89.9 2.8 34.8
61.6 11.9 58.0 1.5 0.7 0.3 35.5
68.1 13.0 60.6 3.2 0.8 0.3 43.9
75.3 14.4 62.3 6.5 1.0 0.4 53.6
61.3 13.8 61.0 2.0 0.5 0.3 12.2
69.8 14.6 64.6 3.8 0.7 0.3 49.7
77.9 15.4 70.3 7.5 0.8 0.4 72.2
0.8 136.2 31.1 2.7 11.4 11.1 145.3
0.2 37.5 10.8 1.5 5.3 3.2 44.4
0.4 71.1 18.1 1.8 7.8 5.2 89.4
0.8 112.0 24.5 2.0 10.2 7.3 143.2
0.3 24.1 12.7 1.3 6.1 3.8 35.8
0.6 55.9 20.9 1.7 8.7 5.9 91.0
1.2 105.1 30.3 2.7 12.4 8.6 170.3
184.0 2.1 84.8 5.0 1.4 9.8 14.6 47.9
146.3 0.8 72.9 2.1 0.6 5.3 8.3 26.9
164.9 182.7 136.1 163.0 181.1 1.5 1.9 1.1 1.6 2.0 82.2 84.2 72.6 81.9 85.6 3.2 4.3 1.0 3.0 4.5 0.9 1.1 0.5 0.9 1.4 8.5 17.5 4.5 10.7 17.5 16.4 19.6 14.1 19.0 24.7 50.6 87.5 28.1 82.1 202.5 41 101
SC
59.1 13.3 60.3 1.5 0.5 0.3 29.2
AC CE P
Kernel properties Hardness (Index) Diameter (mm) Weight (g) Flour properties Milling yield (%) Protein content (14% mb) 3 Mixo water absorption (%) 4 Mixo time (min) Glutenin to Gliadin Ratio 5 H_L_GS_Ratio 6 IPP (%) Dough rheology * Elasticity (N) Extensibility (mm) Work to extend (N.mm) Relaxation time (min) 7 F_25 (N) 8 F_100 (N) Compression force (N) Tortilla properties Tortilla diameter Tortilla specific volume (cm3/g) Lightness (L) Flexibility Score Gradient (N/mm) Force (N) Distance (mm) Work (N.mm) N
NU
38.2 2.5 24.1
Min
2
MA
TIA Mean
PT
Max
Population TXE_UVT 2012 Min Mean Max
Variables
1
Total number of lines evaluated from each of the three wheat populations including outliers. Minimum and maximum values from each population. 3 Mixograph water absorption values; tortilla water absorption values are 10 points less. 4 Dough development time is the mixograph peak time. 5 High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio. 6 Insoluble Polymeric Protein content (as a proportion of total proteins) 7 Stress Relaxation Force after 25 Sec. and after 8100 Sec. compression. *Elasticity means the force recorded as resistance to extension in dough extensibility test. 2
21
ACCEPTED MANUSCRIPT Table 2
Standard
Intercept
1
-1.584
8.32
Hardness
1
-0.050
0.03
3.48
Diameter
1
9.734
4.50
Weight
1
-0.496
0.21
Protein
1
-1.137
1
1
7.274
2
1
-9.273
3
1
4 5 6 7
1
IPP ForceD Area F_25
0.06
0.95
4.68
0.03
16878.63
5.82
0.02
0.61
0.42
7.20
0.01
0.32
3.06
5.64
0.02
1442.90
5.39
2.96
0.09
0.00
-0.053
0.04
2.02
0.16
0.95
1
1.585
1.74
0.83
0.36
4.88
1
-0.171
0.07
6.44
0.01
0.84
1
1.354
0.31
19.19
<.0001
3.87
0.029
0.01
7.08
0.01
1.03
NU
SC
0.21
AC CE P
CompForce
0.85
MA
H_L_GS_Ratio
Exp(Est)
TE
GGRatio
Pr > ChiSq
RI
Estimate
Parameter
D
DF
Wald ChiSquare 0.04
PT
Forward selection procedure analysis of maximum likelihood estimates
1
Glutenin to Gliadin Ratio.
2
High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio.
3
Insoluble Polymeric Protein content
4
Dough elasticity is the force recorded as resistance to extension in dough extensibility test.
5
Work to extend dough strips from the dough extensibility test.
6
Stress Relaxation Force after 25 Seconds of dough compression.
7
Maximum compression force recorded from the dough compression test.
22
ACCEPTED MANUSCRIPT Table 3 Backward elimination procedure analysis of maximum likelihood estimates
Estimate
Standard
Intercept
1
-5.300
7.76
Wald ChiSquare 0.47
Hardness
1
-0.052
0.03
3.90
Diameter
1
9.920
4.39
Weight
1
-0.479
Protein
1
-1.075
1
1
7.846
2
1
-9.976
3
1
4 5
F_25 CompForce
0.01
0.05
0.95
5.10
0.02
20339.65
0.20
5.70
0.02
0.62
0.41
6.76
0.01
0.34
2.92
7.24
0.01
2554.50
5.45
3.35
0.07
0.00
-0.185
0.06
8.58
0.00
0.83
1
1.410
0.28
24.92
<.0001
4.10
1
0.028
0.01
6.71
0.01
1.03
SC
RI
0.49
NU
Area
Exp(Est)
MA
H_L_GS_Ratio
Pr > ChiSq
D
GGRatio
PT
DF
Parameter
Glutenin to Gliadin Ratio.
2
High Molecular Weight to Low Molecular Weight Glutenin Subunit Ratio.
3
Work to extend dough strips from the dough extensibility test.
4
Stress Relaxation Force after 25 Seconds of dough compression.
5
Maximum compression force recorded from the dough compression test.
AC CE P
TE
1
23
ACCEPTED MANUSCRIPT Table 4 Stepwise selection procedure analysis of maximum likelihood estimates
Standard Error
Intercept
1
-10.149
3.35
Weight
1
-0.189
0.08
1
1
10.583
2
1
3 4
Wald ChiSquare
PT
Estimate
Exp(Est)
9.18
0.00
0.00
5.99
0.01
0.83
2.39
19.53
<.0001
39446.00
-0.052
0.03
2.30
0.13
0.95
1
-0.219
0.06
11.82
0.00
0.80
1
1.401
0.32
19.39
<.0001
4.06
5
1
0.355
0.24
2.17
0.14
1.43
6
1
0.021
0.01
4.75
0.03
1.02
Area F_25 F_100 Compression force
SC
IPP
MA
GGRatio
RI
Pr > ChiSq
NU
DF
Parameter
Glutenin to Gliadin Ratio.
2
Insoluble Polymeric Protein content
3
Work to extend dough strips from the dough extensibility test.
4
Stress Relaxation Force after 25 Seconds of dough compression.
5
Stress Relaxation Force after 100 Seconds of dough compression.
6
Maximum compression force recorded from the dough compression test.
AC CE P
TE
D
1
24
ACCEPTED MANUSCRIPT Table 5 Classification summary for cross-validation using five nearest neighbors
POOR
Total Priors
GOOD
POOR
20
61
42
19
67.2
32.8
100
68.9
31.2
25
63
88
19
28.4
71.6
100
21.6
66
83
149
61
44.3
55.7
100
0.41
0.59
0.30
Error
Overall Apparent Error rates 0.28
Error
GOOD
POOR
Total Error
61
46
8
54
100
85.2
14.8
100
69
88
13
60
73
78.4
100
17.8
82.2
100
88
149
59
68
127
40.9
59.1
100
46.5
53.5
100
0.41
0.59
0.43
0.57
0.31
0.22
0.15
0.18
0.26
0.17
TE
D
0.33
Total
AC CE P
Rate
PT
41
Total
SC
GOOD
POOR
NU
Type
GOOD
Stepwise selection
MA
From
Backward elimination
RI
Forward selection
25
ACCEPTED MANUSCRIPT Table 6 Actual number of lines and proportions based on all raw data including outliers 3
Good
Poor
Total
21 (47%)
24 (53%)
TXE/UVT
15 (37%)
26 (63%)
TAM1112
36 (36%)
Total
72 (39%)
45
SC
RI
TIA
41
101
115 (61%)
187
NU
64 (64%)
TIA, specialty lines under development for flatbreads; TXE/UVT, advanced lines and varieties
MA
1
2
Population
PT
1
under development as bread wheat; TAM1112, experimental lines from TAM 111 and TAM 112
Good and Poor means number of lines that were suitable, and not suitable for production of
AC CE P
tortillas based on actual data.
TE
2, 3
D
populations.
26
ACCEPTED MANUSCRIPT
PT
Industrial Relevance
RI
Tortillas and other flatbread manufacturers currently use wheat developed for other commodities and rely
SC
on trial and error, and use of various additives to optimize product quality. Genetic development of wheat for these markets is impeded by lack of knowledge of specific grain quality parameters to target. With the
NU
growing demand for clean label and healthy offerings by consumers, the industry is looking for natural ingredients with improved functionality. This work provides the first insight into the specific wheat
AC CE P
TE
D
MA
composition and functional parameters that can predict tortilla quality.
27
ACCEPTED MANUSCRIPT
PT
Highlights
Traditional bread wheat quality tests poor predictors of functionality in tortilla.
-
Multivariate approach used to identify key wheat quality parameters for tortillas.
-
Glutenin-giladin ratio and insoluble polymeric protein major compositional predictors.
-
Dough extensibility and stress relaxation important functional predictors.
-
Stepwise regression model with 83% prediction efficiency developed.
AC CE P
TE
D
MA
NU
SC
RI
-
28