Efficiently tracking selection in a multiparental population: the case of earliness in wheat

Abstract

Multiparental populations are innovative tools for fine mapping large numbers of loci. Here we explored the application of a wheat Multiparent Advanced Generation Inter-Cross (MAGIC) population for QTL mapping. This population was created by 12 generations of free recombination among 60 founder lines, following modification of the mating system from strict selfing to strict outcrossing using the ms1b nuclear male sterility gene. Available parents and a subset of 380 SSD lines of the resulting MAGIC population were phenotyped for earliness and genotyped with the 9K i-Select SNP array and additional markers in candidate genes controlling heading date. We demonstrated that 12 generations of strict outcrossing rapidly and drastically reduced linkage disequilibrium to very low levels even at short map distances and also greatly reduced the population structure exhibited among the parents. We developed a Bayesian method, based on allelic frequency, to estimate the contribution of each parent in the evolved population. To detect loci under selection and estimate selective pressure, we also developed a new method comparing shifts in allelic frequency between the initial and the evolved populations due to both selection and genetic drift with expectations under drift only. This evolutionary approach allowed us to identify 26 genomic areas under selection. Using association tests between flowering time and polymorphisms, 6 of these genomic areas appeared to carry flowering time QTL, 1 of which corresponds to Ppd-D1, a major gene involved in the photoperiod sensitivity. Frequency shifts at 4 of 6 areas were consistent with earlier flowering of the evolved population relative to the initial population. The use of this new outcrossing wheat population, mixing numerous initial parental lines through multiple generations of panmixia, is discussed in terms of power to detect genes under selection and association mapping. Furthermore we provide new statistical methods for use in future analyses of multiparental populations.

Keywords: MPP; Multiparent Advanced Generation Inter-Cross (MAGIC); QTL detection; experimental evolution; multiparental populations; parental contribution; recombinant population; selection detection.

Figure 1

INRA MAGIC population creation scheme. Intensity of shading represents the proportion of Probus in the population.

Figure 2

Left: Estimated population structure in parental lines. Each individual is represented by a vertical line, which is partitioned into two colored segments that represent the individual’s estimated membership fractions in two clusters. Black vertical line separates European lines (mainly green) from non-European lines (mainly red). Right: Projection of individuals onto discriminant axes with groups represented by different colors: six groups related by pedigree.

Figure 3

Genome-wide estimates of contribution for each parent. Maximum credible interval 0.002. Above bars represent the contribution higher than the median contribution while below bars correspond to contributions lower than the median. Parents with the highest contribution are in red and those with the lowest in green.

Figure 4

LD decay (mean and standard deviation every 5 cM) in the evolved population as a function of genetic distance. Interchromosome mean LD is set at an arbitrary distance of 320 cM.

Figure 5

Manhattan plot of the –log₁₀(P-value) of selection tests. The horizontal line represents the significance threshold (FDR correction = 4.2e-4). Markers under significant selection and associated with heading date are represented by triangles. Ppd-D1 is represented with a triangle on chromosome 2D (P-value < 1e-13).

Figure 6

Distributions of heading date (in degree days) estimated by the Bayesian method. Top: Initial population. Bottom: Evolved population. Proportion of individuals in each class: shaded bars indicate photoperiod-insensitive allele (Ppd-D1a) and open bars indicate photoperiod-sensitive allele (Ppd-D1b).

Figure A1

Scheme of the method of parental contribution estimation. (A) The available data, f(X/L) i.e. the allele frequency at each locus X within each parental line L. (B) The unknown parameters to estimate, P(L) i.e. the probability of contribution to the observed population of each parent L. (C) The observations, f(X) i.e. the allele frequency at each locus. In this example we consider 3 parent lines (L1 in grey, L2 in red and L3 in yellow) genotyped at 3 SNPs (A, in green, B in blue and C in pink). We specify a binomial distribution of the allele frequencies with N the total number of individuals sampled in the observed population. Each locus is treated independently and the aim of the algorithm is estimation of genomic contribution of parents to the population.