Introduction

The goal of cultivar development programs is to advance, through breeding, varieties suitable in quality for release to growers. Wheat breeders cross parental lines to make new varieties that hopefully will have superior agronomic traits (i.e. disease resistance and yield) as well as superior, heritable end-use quality (i.e. milling and baking properties). Due to genetic segregation and recombination, the breeding process quickly becomes an exercise in dealing with large numbers of wheat breeding lines and the copious data associated with each line. Making interpretation difficult, overlaid on the genetic, heritable aspects of end-use quality is the influence of environment, which can cause large variation in quality attributes across different locations and growing years.

The goal of wheat end-use quality testing is to gather information on the potential of a variety to function satisfactorily and economically in milling and baking applications. Moreover, the goal is to make sense of the large volume of information, place the information in context and provide perspective and guidance to wheat breeders as to which varieties to advance in breeding programs and which to discard.

The USDA-ARS Wheat Quality Laboratory (WWQL) in Pullman, WA takes two approaches to variety development “triage”. In early generations, F₄ to F₆, the approach to dealing with large numbers is primarily observational and empirical. In later generations, F₇ and beyond, simple, practical statistics are used. The approaches presented here outline the methods used by our laboratory. The methods discussed here have been successful in guiding wheat breeding programs and can be adapted easily to other programs.

Discussion

Environmental Impact

The Pacific Northwest (PNW) geographical area from which the WWQL receives its wheat grain for testing is extraordinarily diverse. Elevations range from <100 m to 2500 m with great variation in temperature and snow cover associated with the change in altitude. Rainfall varies from 12 cm/yr to 75 cm/yr. Even within the state of Washington, rainfall varies drastically. Starting at the eastern border with Idaho, rainfall decreases 1 cm for each 6 km traveled westward, from 60 cm/yr to 12 cm/yr. The changes in environment have a drastic impact on the quality of the grain produced across the spectrum of growing environments. Thus, difficulties arise in interpreting and comparing data from grain derived from any given location with grain from another location.

Due to environmental variation, no sample can be evaluated in isolation, with the exception of comparing test results to specific, pre-existing parameters, as is done in grain trading. Comparisons to specifications are valid, but reveal nothing about heritable genetic differences among varieties. Standards are needed against which to compare a sample and even then, how is the degree of difference determined to be meaningful?

Early Generation Testing

Each year, the WWQL receives 8,000 wheat lines for evaluation for end-use quality. Of these, 5,000 are fully milled and baked. This produces about 750,000 unique pieces of data that need to be analyzed and placed into context. Simple empirical, observational tests are the easiest methods used to address the need to screen great numbers of samples with large amounts of descriptive information.

It has long been known (Finney, 1948) that among hard wheats, there is genetic predisposition to obtain greater loaf volume response for each incremental unit of increased protein content (Fig 1A). Some wheats bake into larger loaves than do others, even at the same protein content, and the incremental increase in loaf volume for each added amount of protein is also greater. As such, for screening large numbers of new wheat lines, the WWQL has created a “protein response line” (Fig 1B) for bread-baking. If a particular variety falls significantly below the response line at any given protein content, the line is removed from the breeding program. The protein response line was constructed from a regression analysis of several hard wheat varieties that were broadly adapted to many environments and that baked good quality bread. As can be seen in Fig 1B, the distribution of points is skewed above the regression line. When the protein response line was first created, it was centered over the scatter of points. As selection pressure was applied to breeding lines, poorly performing lines were removed and better performing lines were retained. As such, more points appear in the area above the protein response line. The WWQL intends to recalculate the line in the near future to place additional pressure on wheats to perform well in bread baking.

Figure 1. A) Response in 100-g loaf volume of various wheat lines to increase in protein content (Finney, 1948); B) Distribution of 4,000 cultivars and experimental wheat lines for flour protein and 100 g loaf volume with WWQL expected protein response line.

It is known that the WWQL’s baking staff can produce loaves of bread using the AACC 10-10B (100 g straight-dough pup loaf) to within 30 cc repeatedly. The 30 cc range was chosen as the “window” to be used for evaluating whether a new line differed significantly from the expected response curve or not. For data provided to wheat breeders, varieties are flagged as being equal to the expected response line: null, if the line is within 30 cc of the expected response; + or – if the line exceeds or falls short of the expected response by between 30 and 60 cc; and +2 or -2 if the line is greater or less than 60 cc of the expected protein response line. The reporting of data with particularly promising or particularly underperforming lines “flagged” makes data inspection and planting decisions easier for the wheat breeders by placing each line in context and allowing rapid examination of results of many lines. This approach is also used for milling parameters, proximate analysis and soft wheat baking analysis (cookie/biscuit and cake).

The problem with this approach is that it is strictly based on experience and observation; there is no true statistical basis. Using this method for breeding triage is effective in reducing the number of samples carried forward into the next breeding generation by 90%. Thus 1,000 new lines submitted in the F₅ generation are reduced to 1 line at the F₈ generation.

Advanced Generation Testing

In advanced generations (>F₇), the need to identify varietal stability across environments becomes as crucial as is end-use quality. Identification of environmental stbility requires comparing the variety’s performance with that of well-known cultivars. To identify sources of variation within U.S. soft wheat marketing classes (Soft White Winter and Club wheat), a study was undertaken to examine 17 varieties over 8 years in 37 locations (not all the same locations in each year), and the response to flour yield (FYELD), Break Flour Yield (BKFYELD) and cookie diameter (CODI) was analyzed through analysis of variance.

Table 1. Analysis of Variance for selected soft wheat quality parameters for variety, year, environment and year by environment interaction. df = degrees of freedom; Mean sq = parameter mean squares; F-value = parameter F-value. Non-significant (p < 0.05) model terms were omitted from model.

As can be seen in Table 1, the year in which varieties were grown had a substantial and significant role in quality trait variation, as did variety. Environment and year, as well as their interaction, played a lesser role. However, there was no significant interaction between variety and year, or variety and environment. The implication is that varieties will rank in the same order, regardless of growing year or growing environment. This observation allows lines to be grown with a common check variety and statistical analyses to be applied since the varieties and check lines do not change their ranked order. A full model of all locations, years, varieties and environments would have provided better information for statistical analysis, but many locations did not appear in each year’s plantings. Thus the data were “hole-y” and statistical analysis would have been misleading. Therefore, a different type of analysis was used, but an analysis that requires careful selection of standard check varieties.

The statistical model construction requires that a standard wheat variety, well-adapted to many environments and possessing good end-use quality, be grown along with the varieties being tested in several locations and over several years. Each set of samples must be processed with the same testing procedures, on the same equipment (especially important for milling quality estimation). Avoiding use of atypical samples is also a necessity. If pre-harvest sprouting, mechanical damage from harvesting or any other condition that would alter the end-use functionality of the samples is present, the sample set should not be included in the analysis unless there is an over riding need. Breeding for pre-harvest sprout resistant wheat would be an example of an occasion in which wheats from locations impacted by sprouting should be included.

In all locations where paired samples exist (the standard, check variety and the experimental variety), analysis using Analysis of Variance (ANOVA) procedure or a general linear models test to determine the Least Standard Difference (LSD) for the mean values is undertaken for the end-use tests as appropriate. A balanced LSD test is exactly equivalent to a paired t-test analysis. Not only does the LSD test examine whether the two wheats are statistically the same or different, but it also provides a measure of the magnitude of the difference: how many LSD intervals separate the mean values, a measure of the magnitude of the difference.

Table 2 shows two examples of this approach. Two experimental varieties, “X” and “Y” are compared to a common, widely adapted check variety, “Stephens”. Clearly, variety “X” ” (appearing in 61 paired site-year comparisons with “Stephens”) is superior to the check. Whereas variety “Y” (appearing in 69 paired site-year comparisons with “Stephens”) is of inferior quality. In this example, variety “X” should be continued in the breeding program, where variety “Y” should be removed.

Table 2. Mean results of Flour Yield (FYELD), Break Flour Yield (BKFYELD) and Cookie Diameter (CODI) for check variety “Stephens” and experimental lines “X” and “Y”, grown as paired comparisons. LSD groups with the same letter were not significantly different at p = 0.05. Mean values that are significantly better than the paired value are in bold font.

As with the empirical testing approach, data are reported to breeders as: null; + or -; +2 or -2, depending on how many LSD intervals the experimental variety’s mean lies from that of the check variety’s mean.

The LSD statistical testing procedure also affords the ability to continually add data to improve the resolution of the paired comparisons. All that is required is to grow the same pair of wheat varieties together, run the same analytical tests and then regenerate the statistical output with the same ANOVA model. As mentioned above, multiple comparisons among varieties are possible, but are generally confounded due to missing paired data points, usually due to lack of complete site-year sample sets.

The LSD test, or paired t-test comparison, provides statistically valid separations of means that can be used to guide breeding programs. If a variety with potential to be released to commercial production is under consideration, it must necessarily demonstrate sufficient adaptation to the wide range of diverse environments found in the PNW of the U.S. This method provides the means to determine whether the variety will perform satisfactorily or not. Producing wheat that is of uniform, excellent end-use quality is the goal of both the wheat breeding programs in the PNW and the WWQL. The statistical management tools discussed here provide a means to reach that goal.

Further information about the WWQL wheat quality testing program can be obtained at:

http://www.wsu.edu/~wwql

References

American Association of Cereal Chemists, Approved Methods 10^th Edition. (2000) Approved Method 10-10B.

Finney, K.F. & Barmore, M.A. (1948). Cereal Chem. 25:291-312.