Study On Methods For Mapping QTLs Underlying Endosperm Traits In Cereals

Cereal grains are mainly composed of endosperms, which are human's staple food containing rich nutritious substances such as starch, protein and lipid. Many endosperm traits are related to grain yield and quality. Therefore, studying the genetic basis of endosperm traits is very important for cereal crop breeding. Endosperm is a triploid tissue. Its genetic mode is more complicated than diploid tissues. For a locus with two different alleles (say, Q and q), there are four possible genotypes (QQQ, QQq, Qqq and qqq) in the endosperm genome. It requires defining three effects, namely, additive effect (a), first dominance effect (d₁) and second dominance effect (d₂. In most studies having been reported, the mapping of quantitative trait loci (QTL) underlying endosperm traits in cereals was conducted based on diploid genetic models and the maternal effect was not considered. In order to analyze the effects of endosperm QTLs more reasonably, some methods for mapping QTLs underlying endosperm traits based on triploid genetic models have been proposed. However, the results of dissecting various QTL effects are still not quite satisfactory.In this study, considering different ways of endosperm trait phenotyping and different sources of marker information, we proposed five experimental designs and corresponding statistical genetic models and methods for interval mapping of endosperm QTLs based on maximum likelihood approach implemented via expectation-maximization algorithm. The feasibility and efficiency of each method were verified by Monte Carlo simulations. The major results are as follows:1. Methods of endosperm QTL mapping based on random hybridization design (Design I) were proposed for endosperm traits phenotyped based on mixed samples of multiple seeds. In this case, absence of maternal effects should be assumed. The basic idea is: plants (or lines) from a population with known marker genotype information are randomly hybridized to generate a population of hybrid lines for endosperm QTL mapping; a mixture of seeds of each hybrid line is measured for the endosperm trait to get the mean of the line; then endosperm QTL mapping and effect estimation is performed using the endosperm trait means of hybrid lines and the marker genotype information of parental plants (or lines). Simulation study shows that the methods can precisely map endosperm QTLs and unbiasedly and efficiently estimate the three effects (additive effect, first dominance effect, second dominance effect) of endosperm QTLs.2. There are two ways to estimate both the direct effects and the maternaleffects of endosperm QTLs if the endosperm traits are phenotyped fromindividual seeds. The first way (named one-stage hierarchical design) is to usethe marker genotype information from maternal plants and the endospermphenotypes of individual offspring seeds. The second way (named two-stagehierarchical design) is to use the marker genotype information from the embryosof individual offspring seeds, in addition to the marker genotype informationfrom maternal plants and the endosperm phenotypes of individual offspring seeds.For the one-stage hierarchical design, we proposed a method for endosperm QTLmapping using marker genotype information of BC₁F₁ plants (consisting of equalnumber of the progeny of F₁×P₁ and F₁×P₂ with F₁ as the female parent) andendosperm phenotypes of corresponding individual BC₁F_1:2 seeds (Design II).For the two-stage hierarchical design, we proposed a method for endosperm QTLmapping using trait information from F_2:3 endosperms and molecular markerinformation from F₂ plants and F_2:3 embryos (Design III). Results of computersimulations indicate that both designs can obtain estimates of all the effects of anendosperm QTL. The direct additive effect and maternal additive effect can beprecisely estimated in the case of large sample size. However, the estimationprecisions of dominance effects are much lower. Comparatively, the resultsobtained by the two-stage hierarchical design were significantly better than thoseobtained by the one-stage hierarchical design, with more precise estimates andsmaller standard deviations for all the parameters of QTLs.