Genetic association mapping

Genetic association mapping is now in widespread use in plants (for review, see Zhu et al., Reference Zhu, Gore, Buckler and Yu2008) and has led to the identification of quantitative trait loci and genes underlying standing variation (e.g. Brown et al., Reference Brown, Rooney, Franks and Kresovich2008; Harjes et al., Reference Harjes, Rocheford, Bai, Brutnell, Kandianis, Sowinski, Stapleton, Vallabhaneni, Williams, Wurtzel, Yan and Buckler2008). It has also been applied to study the genetic architecture associated with long-term artificial selection for kernel oil composition in maize (Laurie et al., Reference Laurie, Chasalow, LeDeaux, McCarroll, Bush, Hauge, Lai, Clark, Rocheford and Dudley2004).

Association mapping works by identifying significant correlations between allelic and trait variation; however, methods are required to control for non-independence or relatedness among samples comprising many association mapping panels (e.g. Yu et al., Reference Yu, Pressoir, Briggs, Bi, Yamasaki, Doebley, McMullen, Gaut, Nielsen, Holland, Kresovich and Buckler2006). The primary advantages of association mapping are that pre-existing lines from within or across breeding programmes may be used without the need to develop new genetic stocks, multiple alleles can be examined at once and, in some cases, very high-resolution mapping can be achieved because of rapid decay in linkage disequilibrium (such as in maize [Remington et al., Reference Remington, Thornsberry, Matsuoka, Wilson, Whitt, Doeblay, Kresovich, Goodman and Buckler2001]). The trade-off for high-resolution mapping panels is that, in the absence of good candidate genes, extremely large numbers of markers across the genome or whole genome sequences are required to identify associations. Other disadvantages to association mapping are that correction for relatedness, required to reduce false positive rates, can lead to a significant loss in power, that rare but potentially important alleles cannot be detected and that, for association panels including very diverse (some unadapted) germplasm, phenological variation can confound variation in the trait(s) of interest.

Analysis of allelic frequency change

The most common method used to study the genetic architecture of response to artificial selection has been through the analysis of shifts in allelic frequencies. Frequencies are estimated from samples taken from different generations of selection, and for each locus assayed, a statistic is used to test for significant departures from genetic drift expectations or directional change. Significant departures are taken as evidence of selection; it is commonly assumed that the influence of migration and mutation is minimal.

Initial molecular genetic studies seeking insight into artificial selection response were conducted in maize, using handfuls of allozyme loci (Brown, Reference Brown1971; Brown and Allard, Reference Brown and Allard1971; Stuber and Moll, Reference Stuber and Moll1972). These studies set the stage for the use of more efficient marker technologies and improved statistical methodologies that have allowed for genome-wide scale analyses and the identification of loci presumably underlying response to selection (e.g. Labate et al., Reference Labate, Lamkey, Lee and Woodman1999; De Koeyer et al., Reference De Koeyer, Phillips and Stuthman2001; Coque and Gallais, Reference Coque and Gallais2006; Wisser et al., Reference Wisser, Murray, Kolkman, Ceballos and Nelson2008).

There are advantages and limitations to studying recurrently selected populations. A key advantage is that the direct products of a breeding population can be studied in which selection has sorted out the most favourable and unfavourable alleles. Relatively rare but important alleles are enriched by selection and detectable by the analysis of allelic frequency shifts. The primary limitation of this approach is that allelic effects on specific traits cannot often be estimated since selection is not typically conducted or achieved for a single trait. Development of the base population used to initiate selection and the intensity of artificial selection can impact the structure of linkage disequilibrium, which, in turn, can affect the mapping resolution for loci associated with selection. The statistical power of detecting loci through the analysis of allelic frequency shifts also needs to be addressed.

Combining genetic association mapping with analysis of allelic frequency change

In an effort to understand response to artificial selection and the adaptation of exotic germplasm populations to new environments, we have designed and are implementing a framework (Fig. 1) that is expected to counterbalance the opposing limitations and exploit the complementary advantages of association mapping and analysis of allelic frequency change. Recurrently selected populations represent germplasm that has been selected and adapted to an environment of interest. Such populations thus offer the potential of identifying relevant alleles on which to exert selection in a breeding programme. In recurrently selected populations, initially, rare but favourable alleles increase in frequency from generation to generation while common but unfavourable alleles reduce in frequency. Thus, across generations, allelic frequencies at loci underlying selection response would be relatively balanced compared with allelic frequencies in any single generation. Although perfect balance is unlikely due to the transient nature of populations undergoing improvement, this multigenerational sampling strategy will generally overcome a major limitation of association mapping, which is the detection of low-frequency variation. In turn, the application of association mapping to genetically characterize artificial selection response overcomes the major limitation of analysis of allelic frequency change, i.e. the inability to resolve trait-specific locus associations. The ultimate outcome of combining these two approaches is a powerful approach to gain information about the map locations of genomic loci associated with trait-specific ‘response’ variation and insights into the population genetic basis for phenotypic change.

Fig. 1

A genetic framework for studying the genetic architecture of response to artificial selection. The figure depicts a sampling scheme in which random samples of individuals are taken from multiple generations of a recurrently selected population (G₀ representing the base population, G_i representing any intermediate generation and G_n representing the last generation). Tissue is collected from the sample of individuals that are also self-pollinated or topcrossed. Genotyping is conducted on the sample of individuals and phenotyping is conducted on the sample of families or topcrosses. In the analysis phase, association mapping is conducted using genotype data and family or topcross mean performance data. With the same genotype data, allelic frequency shifts can be examined at the quantitative trait loci identified by association mapping.

The application of association genetic mapping to any type of population requires consideration of genetic disequilibrium. For most recurrently selected populations, random mating is conducted among selection units. Therefore, gametic phase disequilibrium present in the initial generation or arising due to selection is expected to decay in each generation. This allows unlinked loci to be independently associated with specific traits, despite the fact that selection was exerted; however, family-level structure may need to be accounted for, if the sample of selection units is small or the chance of sampling siblings is high. The level of resolution of association mapping and allelic frequency tests depends on the extent of disequilibrium at linked loci. Recurrently selected populations are expected to have lower levels of linkage disequilibrium than biparental mapping populations and may be as low as diverse germplasm collections, if sufficient generations of random mating were involved in population development (Breseghello and Sorrells, Reference Breseghello and Sorrells2006).

In the proposed framework, genotyping is conducted on random samples of individuals taken from different generations of selection. The same individuals are self-pollinated or crossed with a common tester to produce entries that are evaluated in replicated trials. Entry means are used to represent the phenotypes of the parental individuals from which genotype data were produced, and association mapping is conducted by fitting the genotype data to the phenotype data. With the same genotype data, estimates of allelic frequencies are obtained and examined across generations of selection, particularly at genomic loci identified via association mapping. The two approaches provide complementary tests to examine loci that are responsive to selection.