| Epistasis and genotype×environment interaction exist universally and have played an important role in the studies of plant genetics, evolution and heterosis. They have been testified by a lot of research. However, most of current methods for quantitative trait loci (QTL) mapping can not simultaneously detect epistasis, including QTL by QTL and QTL by environment. The methods proposed by Zhu's group, under the framework of composite interval mapping with mixed linear approach, could identify epistatic effects and genotype×environment interaction effects (Zhu,1998; Wang et al.,1999; Zhu & Weir,1998). And Yi et al. (2003) used stochastic search variable selection to estimate the two kinds of effects mentioned above under the framework of multiple QTL genetic model. However, these methodologies were suitable only for experimental population from the cross between two inbred lines rather than for inbred lines in plants.Until now there have been several strategies for mining genes in pure inbred lines. Yu et al. (2006) proposed an association mapping, which simultaneously controls population structure effects and polygenes in the background. However, it is based on a single QTL genetic model and multiple QTL are not included simultaneously in one model. To overcome this issue, several approaches have been proposed. Recently, Iwata et al. (2007) proposed a Bayesian variable selection method to map multi-QTL. More recently, Zhang et al.(2008)proposed a multi-QTL Haseman-Elston regression and Lii et al.(2009)suggested multi-QTL in silico mapping. These methods mentioned above were good for mapping QTL in inbred lines, but QTL epistasis and QTL by environment interaction are not present in their genetic models.Novel alleles, gene interaction and QTL-by-environment interaction play an important part in the growth and development in plants. It is crucial for crop breeding including molecular breeding to mine novel allele responsible for important agronomic characters, such as high-yield and disease resistance. Although the methods mentioned above could detect QTL well in inbred lines, they could not mine novel alleles. To make high use of crop germplasm resource, it is necessary to propose an effective approach to mine novel allele in inbred lines. To overcome the shortcomings in the exited methods, in this paper multi-factor analysis of variance (ANOVA) were used to mine novel allele and to detect the epistatic and genotype×environment interaction in inbred lines. The proposed approach was verified by a series of Monte Carlo simulation experiments and real data analysis. The main results and conclusions are as follows:1) A new approach under the framework of multi-QTL genetic model with genetic and statistical hypothesises, along with penalized maximum likelihood (PML) method for parameter estimation, was proposed to mine novel allele in inbred lines, and confirmed by a series of Monte Carlo simulation experiments. Results showed that:1) the power in the detection of QTL for the new method was higher than that for the single factor ANOVA, especially for the QTL with low heritability. The estimates for the effects and the positions of QTL were closer to the corresponding true values for the new method than for the single factor ANOVA, and the false positive rate (FPR) for the new method were all less than 6%o, which were much less than that obtained from single factor ANOVA.2) The capacity of the new technique to accurately estimate parameters was examined for a range of scenarios, i.e., various sample sizes, allelic number and allelic distribution. The power in the detection of QTL and the accuracy of the estimates for the positions and the effects of QTL increase, as the sample size increases, the allelic number decreases or the skewness of allelic distribution increases.2) A new approach under the framework of multi-QTL+environment+QTL-by-environment interaction genetic model with genetic and statistical hypothesises, along with penalized maximum likelihood (PML) method for parameter estimation, was proposed and confirmed by a series of Monte Carlo simulation experiments. Results showed that the power in the detection of QTL, even if with low heritability, was always high, and the estimates for the positions and the effects of QTL are close to the corresponding true values. As the QTL heritability increases, in addition, the power and the accuracy of the estimates for the positions and the effects of QTL increase and the standard deviation for the estimates of residual variance decreases.3) A new approach under the framework of multi-QTL+environment+QTL-by-environment interaction+QTL-by-QTL interaction genetic model with genetic and statistical hypothesises, along with penalized maximum likelihood (PML) method for parameter estimation, was proposed and confirmed by one Monte Carlo simulation experiment. Results showed that the power in the detection of epistatic QTL, with the heritability of 2%, was relatively low, but the power in the detection of epistatic QTL with the heritability of 5% reached 79.5%, the estimates of the effects and the positions of epistatic QTL were close to their true values, and the FPR is 0.59%o.4) Two real datasets from eighty-one cultivars in cotton and two hundreds forty-four F2:3 and F2:4 families in soybean were used to testify the new approaches above. Results showed that four main-effect QTL, one environment and two QTL-by-environment interactions for growth rate in cotton were identified; and fifteen main-effect QTL, three environmental effects and four QTL-by-environment interactions for seed shape traits (seed length, width and thickness) were detected, among these QTL, fourteen QTL are consistent with those detected by joint analysis of all markers. |
