Unveiling the predominance of Saccharum spontaneum alleles for resistance to orange rust in sugarcane using genome-wide association.

KEY MESSAGE: Six QTLs of resistance to sugarcane orange rust were identified in modern interspecific hybrids by GWAS. For five of them, the resistance alleles originated from S. spontaneum. Altogether, they efficiently predict disease resistance. Sugarcane orange rust (SOR) is a threatening emerging disease in many sugarcane industries worldwide. Improving the genetic resistance of commercial cultivars remains the most promising solution to control this disease. In this study, an association panel of 568 modern interspecific sugarcane hybrids (Saccharum officinarum x S. spontaneum) from Réunion's breeding program was evaluated for its resistance to SOR under natural conditions of infection. Two genome-wide association studies (GWAS) were conducted between disease reactions and 183,842 single nucleotide polymorphism (SNP) markers obtained by targeted genotyping-by-sequencing. Five resistance quantitative trait loci (QTLs), named Oru1, Oru2, Oru3, Oru4 and Oru5, were identified using a single-locus GWAS (SL-GWAS). These five QTLs all originated from the species S. spontaneum. A multi-locus GWAS (ML-GWAS) uncovered an additional but less significant resistance QTL named Oru6, which originated from S. officinarum. All six QTLs had a moderate to major phenotypic effect on disease resistance. Prediction accuracy estimated with linear regression models based on each of the five QTLs identified by SL-GWAS was between 0.16-0.41. Altogether, these five QTLs provided a relatively high prediction accuracy of 0.60. In comparison, accuracies obtained with six genome-wide prediction models (i.e., GBLUP, Bayes-A, Bayes-B, Bayes-C, Bayesian Lasso and RKHS) reached only 0.65. The good prediction accuracy of disease resistance provided by the QTLs and the predominant S. spontaneum origin of their resistance alleles pave the way for effective marker-assisted breeding strategies.

Detecting directional and non-directional epistasis in bi-parental populations using genomic data.

Epistasis, commonly defined as interaction effects between alleles of different loci, is an important genetic component of the variation of phenotypic traits in natural and breeding populations. In addition to its impact on variance, epistasis can also affect the expected performance of a population and is then referred to as directional epistasis. Before the advent of genomic data, the existence of epistasis (both directional and non-directional) was investigated based on complex and expensive mating schemes involving several generations evaluated for a trait of interest. In this study, we propose a methodology to detect the presence of epistasis based on simple inbred biparental populations, both genotyped and phenotyped, ideally along with their parents. Thanks to genomic data, parental proportions as well as shared parental proportions between inbred individuals can be estimated. They allow the evaluation of epistasis through a test of the expected performance for directional epistasis or the variance of genetic values. This methodology was applied to two large multiparental populations, i.e. the American maize and soybean nested association mapping populations, evaluated for different traits. Results showed significant epistasis, especially for the test of directional epistasis, e.g. the increase in anthesis to silking interval observed in most maize inbred progenies or the decrease in grain yield observed in several soybean inbred progenies. In general, the effects detected suggested that shuffling allelic associations of both elite parents had a detrimental effect on the performance of their progeny. This methodology is implemented in the EpiTest R-package and can be applied to any bi/multiparental inbred population evaluated for a trait of interest.

Building a Calibration Set for Genomic Prediction, Characteristics to Be Considered, and Optimization Approaches.

The efficiency of genomic selection strongly depends on the prediction accuracy of the genetic merit of candidates. Numerous papers have shown that the composition of the calibration set is a key contributor to prediction accuracy. A poorly defined calibration set can result in low accuracies, whereas an optimized one can considerably increase accuracy compared to random sampling, for a same size. Alternatively, optimizing the calibration set can be a way of decreasing the costs of phenotyping by enabling similar levels of accuracy compared to random sampling but with fewer phenotypic units. We present here the different factors that have to be considered when designing a calibration set, and review the different criteria proposed in the literature. We classified these criteria into two groups: model-free criteria based on relatedness, and criteria derived from the linear mixed model. We introduce criteria targeting specific prediction objectives including the prediction of highly diverse panels, biparental families, or hybrids. We also review different ways of updating the calibration set, and different procedures for optimizing phenotyping experimental designs.

Assessment of genomic prediction reliability and optimization of experimental designs in multi-environment trials.

KEY MESSAGE: New forms of the coefficient of determination can help to forecast the accuracy of genomic prediction and optimize experimental designs in multi-environment trials with genotype-by-environment interactions. In multi-environment trials, the relative performance of genotypes may vary depending on the environmental conditions, and this phenomenon is commonly referred to as genotype-by-environment interaction (G[Formula: see text]E). With genomic prediction, G[Formula: see text]E can be accounted for by modeling the genetic covariance between trials, even when the overall experimental design is highly unbalanced between trials, thanks to the genomic relationship between genotypes. In this study, we propose new forms of the coefficient of determination (CD, i.e., the expected model-based square correlation between a genetic value and its corresponding prediction) that can be used to forecast the genomic prediction reliability of genotypes, both for their trial-specific performance and their mean performance. As the expected prediction reliability based on these new CD criteria is generally a good approximation of the observed reliability, we demonstrate that they can be used to optimize multi-environment trials in the presence of G[Formula: see text]E. In addition, this reliability may be highly variable between genotypes, especially in unbalanced designs with complex pedigree relationships between genotypes. Therefore, it can be useful for breeders to assess it before selecting genotypes based on their predicted genetic values. Using a wheat population evaluated both for simulated and phenology traits, and two maize populations evaluated for grain yield, we illustrate this approach and confirm the value of our new CD criteria.

Genomic prediction and training set optimization in a structured Mediterranean oat population.

KEY MESSAGE: The strong genetic structure observed in Mediterranean oats affects the predictive ability of genomic prediction as well as the performance of training set optimization methods. In this study, we investigated the efficiency of genomic prediction and training set optimization in a highly structured population of cultivars and landraces of cultivated oat (Avena sativa) from the Mediterranean basin, including white (subsp. sativa) and red (subsp. byzantina) oats, genotyped using genotype-by-sequencing markers and evaluated for agronomic traits in Southern Spain. For most traits, the predictive abilities were moderate to high with little differences between models, except for biomass for which Bayes-B showed a substantial gain compared to other models. The consistency between the structure of the training population and the population to be predicted was key to the predictive ability of genomic predictions. The predictive ability of inter-subspecies predictions was indeed much lower than that of intra-subspecies predictions for all traits. Regarding training set optimization, the linear mixed model optimization criteria (prediction error variance (PEVmean) and coefficient of determination (CDmean)) performed better than the heuristic approach "partitioning around medoids," even under high population structure. The superiority of CDmean and PEVmean could be explained by their ability to adapt the representation of each genetic group according to those represented in the population to be predicted. These results represent an important step towards the implementation of genomic prediction in oat breeding programs and address important issues faced by the genomic prediction community regarding population structure and training set optimization.

TrainSel: An R Package for Selection of Training Populations.

A major barrier to the wider use of supervised learning in emerging applications, such as genomic selection, is the lack of sufficient and representative labeled data to train prediction models. The amount and quality of labeled training data in many applications is usually limited and therefore careful selection of the training examples to be labeled can be useful for improving the accuracies in predictive learning tasks. In this paper, we present an R package, TrainSel, which provides flexible, efficient, and easy-to-use tools that can be used for the selection of training populations (STP). We illustrate its use, performance, and potentials in four different supervised learning applications within and outside of the plant breeding area.

Accounting for Group-Specific Allele Effects and Admixture in Genomic Predictions: Theory and Experimental Evaluation in Maize.

Populations structured into genetic groups may display group-specific linkage disequilibrium, mutations, and/or interactions between quantitative trait loci and the genetic background. These factors lead to heterogeneous marker effects affecting the efficiency of genomic prediction, especially for admixed individuals. Such individuals have a genome that is a mosaic of chromosome blocks from different origins, and may be of interest to combine favorable group-specific characteristics. We developed two genomic prediction models adapted to the prediction of admixed individuals in presence of heterogeneous marker effects: multigroup admixed genomic best linear unbiased prediction random individual (MAGBLUP-RI), modeling the ancestry of alleles; and multigroup admixed genomic best linear unbiased prediction random allele effect (MAGBLUP-RAE), modeling group-specific distributions of allele effects. MAGBLUP-RI can estimate the segregation variance generated by admixture while MAGBLUP-RAE can disentangle the variability that is due to main allele effects from the variability that is due to group-specific deviation allele effects. Both models were evaluated for their genomic prediction accuracy using a maize panel including lines from the Dent and Flint groups, along with admixed individuals. Based on simulated traits, both models proved their efficiency to improve genomic prediction accuracy compared to standard GBLUP models. For real traits, a clear gain was observed at low marker densities whereas it became limited at high marker densities. The interest of including admixed individuals in multigroup training sets was confirmed using simulated traits, but was variable using real traits. Both MAGBLUP models and admixed individuals are of interest whenever group-specific SNP allele effects exist.

Disentangling group specific QTL allele effects from genetic background epistasis using admixed individuals in GWAS: An application to maize flowering.

When handling a structured population in association mapping, group-specific allele effects may be observed at quantitative trait loci (QTLs) for several reasons: (i) a different linkage disequilibrium (LD) between SNPs and QTLs across groups, (ii) group-specific genetic mutations in QTL regions, and/or (iii) epistatic interactions between QTLs and other loci that have differentiated allele frequencies between groups. We present here a new genome-wide association (GWAS) approach to identify QTLs exhibiting such group-specific allele effects. We developed genetic materials including admixed progeny from different genetic groups with known genome-wide ancestries (local admixture). A dedicated statistical methodology was developed to analyze pure and admixed individuals jointly, allowing one to disentangle the factors causing the heterogeneity of allele effects across groups. This approach was applied to maize by developing an inbred "Flint-Dent" panel including admixed individuals that was evaluated for flowering time. Several associations were detected revealing a wide range of configurations of allele effects, both at known flowering QTLs (Vgt1, Vgt2 and Vgt3) and new loci. We found several QTLs whose effect depended on the group ancestry of alleles while others interacted with the genetic background. Our GWAS approach provides useful information on the stability of QTL effects across genetic groups and can be applied to a wide range of species.

Genomic selection efficiency and a priori estimation of accuracy in a structured dent maize panel.

KEY MESSAGE: Population structure affects genomic selection efficiency as well as the ability to forecast accuracy using standard GBLUP. Genomic prediction models usually assume that the individuals used for calibration belong to the same population as those to be predicted. Most of the a priori indicators of precision, such as the coefficient of determination (CD), were derived from those same models. But genetic structure is a common feature in plant species, and it may impact genomic selection efficiency and the ability to forecast prediction accuracy. We investigated the impact of genetic structure in a dent maize panel ("Amaizing Dent") using different scenarios including within- or across-group predictions. For a given training set size, the best accuracies were achieved when predicting individuals using a model calibrated on the same genetic group. Nevertheless, a diverse training set representing all the groups had a certain predictive efficiency for all the validation sets, and adding extra-group individuals was almost always beneficial. It underlines the potential of such a generic training set for dent maize genomic selection applications. Alternative prediction models, taking genetic structure explicitly into account, did not improve the prediction accuracy compared to GBLUP. We also investigated the ability of different indicators of precision to forecast accuracy in the within- or across-group scenarios. There was a global encouraging trend of the CD to differentiate scenarios, although there were specific combinations of target populations and traits where the efficiency of this indicator proved to be null. One hypothesis to explain such erratic performances is the impact of genetic structure through group-specific allele diversity at QTLs rather than group-specific allele effects.