Relationship of Bread Quality to Kernel, Flour, and Dough Properties

Cereal Chem. 85(l):82-91 This study measured the relationship between bread quality and 49 hard red spring (HRS) or 48 hard red winter (HRW) grain, flour, and dough quality characteristics. The estimated bread quality attributes included loaf volume, bake mix time, bake water absorption, and crumb grain score. The best-fit models for loaf volume, bake mix time, and water absorption had R2 values of 0.78-0.93 with five to eight variables. Crumb grain score was not well estimated, and had R2 values 0.60. For loaf volume models, grain or flour protein content was the most important parameter included. Bake water absorption was best estimated when using mixograph water absorption, and flour or grain protein content. Bread quality is difficult to predict from kernel, flour, or dough characteristics. In many wheat breeding programs, thousands of new lines are tested every year to find high-quality wheat for breadmaking. Most early generation lines are produced in a very limited quantity which does not allow baking tests to be conducted. Therefore, the ability to estimate bread quality using limited sample sizes will be highly beneficial to wheat breeding programs. In addition, if bread quality could be rapidly predicted from grain or flour, millers and bakers could adjust their processes to maximize profits and give consumers a consistently highquality product. Various researchers have attempted to predict bread quality by combining measurements made from grain, flour, or dough and combining them into prediction models. Millar (2003) used stepwise regression to develop an equation to predict loaf volume (800-g loaves, n = 181). This equation included glutenin quantity, % gliadins, flour color grade, protein content, glutenin elastic modulus, farinograph water absorption, particle size index, moisture content, and the ratio of HMW glutenins to LMW glutenins. This equation gave a standard error (SE) of 161 cm' for loaf volume and a R2 = 0.39. Their equation showed that glutenin mass, protein content, and the ratio of HMW to LMW glutenins had a positive influence on loaf volume, and that flour color grade and particle size index had a negative influence on loaf volume. In color measurement Millar (2003), a higher color grade indicated more ash in the flour. However, farinograph water absorption, glutenin elastic modulus, and gliadins variables had coefficients that were opposite of what was expected, and including moisture content was difficult to explain. Their results indicate the difficulty in developing a single model to predict baking performance. Parameters not significantly influencing loaf volume were falling number, starch damage, and levels of albumins and globulins. USDA ARS, Grain Marketing and Production Research Center, Engineering Research Unit, 1515 College Avenue, Manhattan, KS 66502. Mention of trade names or commercial products is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. 2 Coesnding author. Phone: 785-776-2753. Fax: 785-537-5550. E-mail address: [email protected] USDA. Grain Inspection, Packers, and Stockyards Administration, Federal Grain Inspection Service, Kansas City, MO 64163. Kansas State University, Dept. Grain Science and Industry, Manhattan, KS 66506. USDA ARS, Grain Marketing and Production Research Center, Grain Quality and Structure Research Unit, 1515 College Avenue, Manhattan, KS 66502. doi:l 0.1 094/CCHEM-85-1 -0082 This article is in the public domain and not copyrightable. It may be freely reprinted with customary crediting of the source. AACC International, Inc., 2008. 82 CEREAL CHEMISTRY Bake water absorption models could generally be improved by including farinograph, mixograph, or alveograph measurements. Bake mix time was estimated best when using mixograph mix time, and models could be improved by including glutenin data. When the data set was divided into calibration and prediction sets, the loaf volume and bake mix time models still looked promising for screening samples. When including only variables that could be rapidly measured (protein content, test weight, single kernel moisture content, single kernel diameter, single kernel hardness, bulk moisture content, and dark hard and vitreous kernels), only loaf volume could be predicted with accuracies adequate for screening samples. Lee et al (2006) predicted loaf volume in a study using blends of the hard white wheat cultivars Betty and Trego and achieved an R2 = 0.70. The baked loaves were from lOO-g flour samples (n = 189). Their prediction equation included grain protein content, hardness index, mixograph water absorption and peak height, and break flour extraction. All variables were positively correlated to loaf volume. Although the prediction equation had a high coefficient of determination, all samples were blends of the two original samples and there was limited variability in the sample set. Andersson et al (1994) predicted loaf volume using partial least squares regression models that included bulk density, flour and grain protein content, flour and grain falling number, flour and grain moisture content, ash content, flour yield, and farinograph and extensigraph measurements from 100 samples. Andersson et al (1994) baked loaves from 750-950 g of flour and showed that loaf volume was consistently influenced by grain and flour protein content, farinograph dough development, stability, and breakdown, and extensigraph area, peak height, and length. Flour protein content explained =50% of the variation in loaf volume, while the addition of all variables explained 65.4% of the variation. The models of Andersson et al (1994) predicted loaf volume with a standard error of prediction of 75 cm3. Flour protein content was shown by Graybosch et al (1993) to be the primary factor contributing to dough strength and loaf characteristics. However, they noted that no single biochemical component explained >41% of the variation in bread quality and that prediction models would require measurement of numerous components. The canonical analysis by Graybosch et al (1993) showed that loaf volume, bake water absorption. mix time, and texture were influenced by flour protein content, gliadin, glutenin, water-soluble pentosan, LMW residue protein, salt-water soluble protein, total free lipid, and total free polar lipid contents. Their analysis did not combine measurements into prediction models. Haley et al (1999) developed a relational database where a user could access quality measurements made on samples obtained from the hard winter wheat regional testing program. A baking score could be obtained after the user input the weights assigned to quality measurements such as flour protein, mix time, bake water absorption, and loaf volume. Other researchers (Shuey et al 1975; Dick and Shuey 1976; Nolte et al 1985; Morris and Raykowski 1993) used a similar approach for evaluating the quality of other wheat classes, but databases only allowed samples to be compared with a target or to each other, and quality predictions were not made. The USDA, ARS, Grain Marketing and Production Research Center (GMPRC), Hard Winter Wheat Quality Laboratory, developed an equation to assign a hard winter wheat bake quality score using mixograph water absorption, loaf volume, crumb color, crumb grain, crumb texture, and mixograph mix time using a scale of 0-6 (Anonymous 2005a). A hard spring wheat marketing score was developed to facilitate a better understanding of wheat quality in marketing systems (Anonymous 2005h). The 1 score was determined by using test weight, thousand kernel weight (TKW), falling number, protein content, and ash content. ,The resulting score was meant to provide information to buyers who are purchasing wheat with a specific end-use in mind. The three examples mentioned above were not quality prediction systems of the final wheat products, but rather, they compared the quality of a given wheat cultivar to others provided by the same nursery or by the Wheat Quality Council to make decisions on the status of each line for further breeding stages. Some researchers have attempted to explain some variations in bread quality using various grain, flour, or dough traits and have tried to further develop predictive models. The research reported herein utilizes additional grain, flour, and dough properties that are combined into models to estimate bread quality. The specific objective of this research was to estimate bake water absorption, bake mix time, crumb grain score, and loaf volume from best-fit models that were developed using 5() measures of grain, flour, and dough quality parameters. MATERIALS AND METHODS Wheat Samples One hundred HRW and 100 HRS wheat samples 0 kg each) from the 2002 and 2003 crop year were selected primarily based on protein content and were expected to result in a wide range of bread quality. Two HRS wheat samples were discarded from the sample set due to insect infestation. Maghirang et al (2006) reported the average and standard deviation of the quality factors for these samples and gave details on their source. Samples were obtained from the USDA Grain Inspection, Packers, and Stockyards Administration (GIPSA), Federal Grain Inspection Service (FGIS), Kansas City, MO. Wheat Quality Analysis There were a total of 48 HRW and 49 HRS wheat grain, flour, and dough characteristics measured as described by Maghirang et al (2006). Standard methods were used whenever an approved method was available. Seven whole-grain quality characteristics were measured including test weight (Approved Method 55-10, AACC International 2000), protein content (AACC Approved Method 39-25), moisture content as measured by the DICKEYjohn GAC (Auburn, IL), TKW, single-kernel hardness (AACC Approved Method 55-31), single-kernel moisture content, and mean kernel diameter using SKCS. In addition, percentage of dark hard and vitreous kernels was measured on HRS wheat samples. Milling and flour quality indicators measured (28 total) were flour yield (Approved Method 26-1A, AACC International 2000), wheat and flour ash content (AACC Approved Method 0801), flour protein content (AACC Approved Method 39-11), L*, a*, b* using a colorimeter (CR-300 Minolta, Osaka, Japan), geometric mean diameters of flour particles and starch granules using a laser light-scattering particle-size instrument (Beckman/Coulter 13 320, Fullerton, CA) equipped with Beckman/Coulter application software (v.4.21), polyphenol oxidase (PPO) content (AACC Approved Method 22-85), falling number (AACC Approved Method 56-81B), SDS sedimentation volume (AACC Approved Method 56-70), total wet gluten content and gluten index (AACC Approved Method 38-12), percentage and mass of insoluble, soluble, and total glutenin proteins, percentage and mass of soluble gliadin proteins, the ratios of insoluble glutenins (%)/soluble glutenins (%), insoluble glutenins (%)/total glutenins (%), and soluble gliadins (%)/total glutenins (%), free lipids (%), polar lipids (%), and nonpolar lipids (%). Protein characterization used the procedure outlined by Bean et al (1998). Lipids were measured as described by Chung et al (1980), Ohm and Chung (1999), and Hubbard et al (2004). The 13 dough properties were evaluated using the mixograph (Approved Method 54-40A, AACC lnternatibnal 2000), farinograph (AACC Approved Method 54-21), and alveograph (AACC Approved Method 54-30A). Parameters measured by the mixograph were water absorption, mix time, and mixing tolerance. The parameters measured by the farinograph were water absorption, development time, stability, tolerance, and quality number. The parameters measured by the alveograph were peak height, length, swelling index, work, and configuration ratio. Four breadmaking quality parameters for the pup loaf (100 g of flour) straight-dough procedures were measured: bake water absorption, bake mix time, crumb grain score, and loaf volume (Approved Method 10-1013, AACC International 2000). The determination of optimum mixing time is as described by Finney (1984). Data Analyses Maghirang et al (2006) showed that HRS wheat quality was significantly different from HRW wheat, even at similar protein content ranges. The two classes were modeled separately in this research. Regression models were developed using the Statistical Analysis System (SAS Institute, Cary, NC) Proc REG procedure with the MAXR selection. The Mallows Cp statistic (Martens and Naes 1989) was also used to evaluate models. A Cp value larger than the number of variables in the model indicates that the model will have a bias and Cp values should be about equal to the number of variables in the model or be negative. In addition, all combinations of variables that gave predictions with R2 > 0.70 were calculated using the PROC REG procedures with the RSQUARE selection. Four HRW samples had missing data for TKW, two samples had missing alveograph measurements, and one sample had missing flour particle size data. Thus, 93 samples were included in the HRW models. For HRS wheat models, six samples had missing TKW data, one had missing protein quality data, and one had missing alveograph data. Those samples were eliminated from the analyses, resulting in data from 90 HRS wheat samples used in the models. Additional models were developed using arameters that could be measured rapidly using instrumentation in field locations. These parametdrs included test weight, grain protein content, single kernel hardness and diameter, moisture content, and TKW for HRW and HRS wheat. The percentage of dark hard and vitreous kernels was included in the HRS wheat models. Dowell et al (2006) attempted to rapidly predict all flour, dough, and bread quality measurements used in this study by NIRS of whole kernels but none of these parameters could be predicted independently of their relationship with protein content. The HRS samples had higher average protein content than the HRW samples (14.6 vs. 12.6%) (Maghirang et al 2006). In an attempt to remove any bias caused by protein content differences from our analyses, additional models were developed which included only samples in the 11.4-15.8% protein content range. These models included 75 HRS and 73 HRW samples. The INFLUENCE option in the Proc REG procedure was used to measure the influence of each sample on the prediction models by calculating the studentized residual (RSTUDENT). Samples with an RSTU DENT greater than two were deleted from the analysis and compared with results with all samples included in the models. Calibration models were developed from 80% of the data and used to predict the remaining samples. The prediction set was selected by removing every fifth sample from the data set. This resulted in 74 calibration set samples and 19 prediction set samples for HRW wheat, and 72 calibration set samples and 18 prediction set samples for HRS wheat. Vol. 85, No. 1, 2008 83