Potential of genomic selection in rapeseed (Brassica napus L.) breeding

Genomic selection employs genome‐wide marker data to predict genomic breeding values. In this study, a population consisting of 391 lines of elite winter oilseed rape derived from nine families was used to evaluate the prospects of genomic selection in rapeseed breeding. All lines have been phenotyped for six morphological, quality‐ and yield‐related traits and genotyped with genome‐wide SNP markers. We used ridge regression best linear unbiased prediction in combination with cross‐validation and obtained medium to high prediction accuracies for the studied traits. Our results illustrate that among‐family variance contributes to the prediction accuracy and can lead to an overestimation of the prospects of genomic selection within single segregating families. We also tested a scenario where estimation of effects was carried out without individuals from the family in which breeding values were predicted, which yielded lower but nevertheless attractive prediction accuracies. Taken together, our results suggest that genomic selection can be a valuable genomic approach for complex agronomic traits towards a knowledge‐based breeding in rapeseed.     

Choice of shrinkage parameter and prediction of genomic breeding values in elite maize breeding populations

Genomic selection (GS) is a promising alternative to marker‐assisted selection particularly for quantitative traits. In this study, we examined the prediction accuracy of genomic breeding values by using ridge regression best linear unbiased prediction in combination with fivefold cross‐validation based on empirical data of a commercial maize breeding programme. The empirical data is composed of 930 testcross progenies derived from 11 segregating families evaluated at six environments for grain yield and grain moisture. Accuracy to predict genomic breeding values was affected by the choice of the shrinkage parameter λ², by unbalanced family size, by size of the training population and to a lower extent by the number of markers. Accuracy of genomic breeding values was high suggesting that the selection gain can be improved implementing GS in elite maize breeding programmes.     

Genomic prediction of grain yield in commercial Finnish oat (Avena sativa) and barley (Hordeum vulgare) breeding programmes

Genomic selection has been adopted in many plant breeding programmes. In this paper, we cover some aspects of information necessary before starting genomic selection. Spring oat and barley breeding data sets from commercial breeding programmes were studied using single, multitrait and trait-assisted models for predicting grain yield. Heritabilities were higher when estimated using multitrait models compared to single-trait models. However, no corresponding increase in prediction accuracy was observed in a cross-validation scenario. On the other hand, forward prediction showed a slight, but not significant, increase in accuracy of genomic estimated breeding values for breeding cohorts when a multitrait model was applied. When a correlated trait was used in a trait-assisted model, on average the accuracies increased by 9%–14% for oat and by 11%–28% for barley compared with a single-trait model. Overall, accuracies in forward validation varied between breeding cohorts and years for grain yield. Forward prediction accuracies for multiple cohorts and multiple years’ data are reported for oat for the first time.     

Genome-enabled prediction models for black tea (Camellia sinensis) quality and drought tolerance traits

Genomic selection in tea plant (Camellia sinensis) breeding has the potential to accelerate efficiency of choosing parents with desirable traits at the seedling stage. The study evaluated different genome-enabled prediction models for black tea quality and drought tolerance traits in discovery and validation populations. The discovery population comprised of two segregating tea populations (TRFK St. 504 and TRFK St. 524) with 255 F₁ progeny and 56 individual tea cultivars in validation population genotyped using 1,421 DArTseq markers. Twofold cross-validation was used for training the prediction models in the discovery population on eight different phenotypic traits. The best prediction models in the discovery population were consequently fitted to the validation population. Of all the four model-based prediction approaches, putative QTLs (Quantitative Trait Loci) + annotated proteins + KEGG (Kyoto Encyclopaedia of Genes and Genomes) pathway-based prediction approach showed more robustness. The findings have for the first time opened up a new avenue for future application of genomic selection in tea breeding.     

Prospects for genomic selection in forage plant species

Genomic selection (GS) is a powerful method for exploitation of DNA sequence polymorphisms in breeding improvement, through the prediction of breeding values based on all markers distributed genome‐wide. Forage grasses and legumes provide important targets for GS implementation, as many key traits are difficult or expensive to assess, and are measured late in the breeding cycle. Generic attributes of forage breeding programmes are described, along with status of genomic resources for a representative species group (ryegrasses). Two schemes for implementing GS in ryegrass breeding are described. The first requires relatively little modification of current schemes, but could lead to significant reductions in operating cost. The second scheme would allow two rounds of selection for key agronomic traits within a time period previously required for a single round, potentially leading to doubling of genetic gain rate, but requires a purpose‐designed reference population. In both schemes, the limited extent of linkage disequilibrium (LD), which is the major challenge for GS implementation in ryegrass breeding, is addressed. The strategies also incorporate recent advances in DNA sequencing technology to minimize costs.     

Identifying superior parents in a potato breeding program using cross prediction techniques

Cross prediction techniques were applied to data collected from over 600 hybrid combinations of potato (Solanum tuberosum) and analyzed to determine the potential of using early generation cross prediction techniques to identify superior parental clones. Performance of parental lines based on parameters collected in early generations were compared with the observed frequencies of desirable recombinants with a common parent in the latter stages of a breeding program. Results showed that value of parents, in their ability to produce desirable recombinants in breeding programs, can be predicted using univariate cross prediction techniques. This type of information can be available from early generation progeny trials and could easily be incorporated into a practical potato breeding scheme. This revised version was published online in August 2006 with corrections to the Cover Date.     

Accuracy of within‐ and among‐family genomic prediction in triticale

Genomic prediction has emerged as a powerful genomic tool to assist breeding of complex traits. In this study, we employed a population of 647 triticale doubled haploid lines derived from four families to assess the potential of this approach for triticale breeding. All lines were phenotyped for grain yield, thousand‐kernel weight, biomass yield, plant height, frost tolerance and Fusarium head blight resistance. The obtained prediction accuracies were moderate to high and consisted to varying degrees of within‐ and among‐family variance, in line with the different degrees of phenotypic differences between family means. The prediction accuracy within individual families also varied with the genetic complexity of the traits and was generally highest based on effect estimation with lines from the respective family, whereas the prediction accuracy decreased with decreasing relatedness among the families. Taken together, our results illustrate the potential of genomic prediction to increase selection gain in triticale breeding, but the composition of the training set is of utmost importance, and consequently, the implementation of this approach in applied breeding programmes is not straightforward.     

中国马铃薯品质现状及改良对策

161606黑龙江省克山县（黑龙江省农科院马铃薯研究所）     

Genomic selection and enablers for agronomic traits in maize (Zea mays): A review

Maize is a commodity crop providing millions of people with food, feed, industrial raw material and economic opportunities. However, maize yields in Africa are relatively stagnant and low, at a mean of 1.7 t ha⁻¹ compared with the global average of 6 t ha⁻¹. The yield gap can be narrowed with accelerated and precision breeding strategies that are required to develop and deploy high-yielding and climate-resilient maize varieties. Genomic and phenotypic selections are complementary methods that offer opportunities for the speedy choice of contrasting parents and derived progenies for hybrid breeding and commercialization. Genomic selection (GS) will shorten the crop breeding cycle by identifying and tracking desirable genotypes and aid the timeous commercialization of market-preferred varieties. The technology, however, has not yet been fully embraced by most public and private breeding programmes, notably in Africa. This review aims to present the importance, current status, challenges and opportunities of GS to accelerate genetic gains for economic traits to speed up the breeding of high-yielding maize varieties. The first section summarizes genomic selection and the contemporary phenotypic selection and phenotyping platforms as a foundation for GS and trait integration in maize. This is followed by highlights on the reported genetic gains and progress through phenotypic selection and GS for grain yield and yield components. Training population development, genetic design and statistical models used in GS in maize breeding are discussed. Lastly, the review summarizes the challenges of GS, including prediction accuracy, and integrates GS with speed breeding, doubled haploid breeding and genome editing technologies to increase breeding efficiency and accelerate cultivar release. The paper will guide breeders in selection and trait introgression using GS to develop cultivars preferred by the marketplace.     

Hybrid breeding for biomass yield in winter triticale: II. Combining ability and hybrid prediction

Accurate hybrid prediction and knowledge about the relative contribution of general (GCA) and specific combining ability (SCA) are of utmost importance for efficient hybrid breeding. We therefore evaluated 91 triticale single-cross hybrids in field trials at seven environments for plant height, heading time, fresh biomass, dry matter content and dry biomass. Fresh and dry biomass showed the highest proportion (23%) of variance due to SCA. Prediction accuracies based on GCA were slightly higher than based on mid-parent values. Utilizing parental kinship information yielded the highest prediction accuracies when both parental lines have been tested in other hybrid combinations, but still moderate-to-low prediction accuracies for two untested parents. Thus, hybrid prediction for biomass traits in triticale is currently promising based on mid-parent values as emphasized by our simulation study, but can be expected to shift to GCA-based prediction with an increasing importance of GCA due to selection in hybrid breeding. Moreover, the performance of potential hybrids between newly developed lines can be predicted with moderate accuracy using genomic relationship information.     

Genomic prediction of the general combining ability of maize lines (Zea mays L.) and the performance of their single crosses

The objective of this study was to assess the effectiveness of genomic selection (GS) on predicting the general combining ability (GCA) of maize lines and the performance of their single crosses. Eight maize lines developed from the different self‐pollination generations of Chalqueño race, along with their 24 single crosses, were evaluated in the field during the years of 2011, 2012 and 2013. Genomic prediction results using genotyping‐by‐sequencing‐based single nucleotide polymorphisms showed that the GCA classification of the parental lines estimated from the SNP information was consistent with the phenotypic classification of the lines evaluated from the field trial data. The prediction accuracy values estimated from the cross‐validation method ranged from 0.49 to 0.61 in the different prediction models. Yield performance of the unevaluated single crosses was predicted based on their SNP information. The total genetic variance of the yield of the single crosses was most explained by the GCA effects. Compared with phenotyping method, GS is a more effective and efficient approach to predict the GCA of maize lines and their hybrid performance.     

Genetic relatedness and the ratio of subpopulation‐common alleles are related in genomic prediction across structured subpopulations in maize

Genomic prediction (GP), which could predict the breeding value of crop plants genotyped with molecular markers, has been carried out in multiple species. Prediction accuracy (PA) of GP depends on various factors, including genetic relatedness and genetic basis. In this study, we examined the rationale for the low PA of GP when the training and validation populations were distinct using 170 temperate inbred lines and 210 tropical and subtropical inbred lines, respectively. All inbred lines were evaluated for 17 traits and genotyped with 550K high‐density markers. The results show that: (a) the influences of heritability and marker number on PA reflected variations in phenotypic variance captured by the genetic information; (b) the low PA of GP when the training and validation populations represent structured subpopulation is related to the ratio of subpopulation‐common alleles (RSCA) and the genetic relatedness between the two subpopulations; (c) RSCA and PA increased with the increase of genetic relatedness, suggesting that these three factors were related. Our findings would provide references when performing GP, and guidance when designing breeding populations.     

Trial heterogeneity and variance models in the genetic evaluation of potato tuber yield

Genetic evaluation aims to identify genotypes with high empirical breeding values (EBVs) for selection as parents. In this study, 2157 potato genotypes were evaluated for tuber yield using 8 years of early‐stage trial data collected from a potato breeding programme. Using linear mixed models, spatial parameters to target greater control of localised spatial heterogeneity within trials were estimated and variance models to account for across‐trial genetic heterogeneity were tested. When spatial components improved model fit, correlations of errors were mostly small and negative for marketable tuber yield (MTY) and total tuber yield (TTY), suggesting the presence of interplot competition in some years. For the analysis of multi‐environment trials, a variance model with a simple correlation structure (with heterogeneous variances) was the most favourable variance structure fitted for TTY and PTY (per cent marketable yield). There was very little difference in model fit when comparing a factor analytic structure of order 2 (FA2) with either FA1 or simple correlation structures for MTY, indicating that simple variance models may be preferable for early‐stage genetic evaluation of potato yield.     

Designing optimal training sets for genomic prediction using adversarial validation with probit regression

Genomic selection (GS) is a disruptive methodology that is revolutionizing animal and plant breeding. However, its practical implementation is challenging since many times there is a mismatch in the distribution of the training and testing sets. Adversarial validation is an approach popular in machine learning to detect and address the difference between the training and testing distributions. For this reason, the adversarial validation method in this research was implemented using probit regression to detect the mismatch in distributions and also to select an optimal training set. We evaluated the proposed method with 14 datasets, and the results were benchmarked regarding of using the whole reference population and simple random samples. We found that the proposed method is effective for detecting the mismatch in distributions and outperformed in prediction accuracy by 11.67% (in terms of mean square error) and by 5.35% (in terms of normalized mean square error) when the whole reference population was used as training sets. Also, in general, this outperformed some existing methods for optimal training designs in the context of GS.     

Replicate allocation to improve selection efficiency in the early stages of a potato breeding scheme

Field data and simulation were used to investigate replication within trials and the allocation of replicates across trial sites using partial replication as an approach to improve the efficiency of early-stage selection in a potato breeding programme. Analysis of potato trial data using linear mixed models, based on four-plant (clonal) plots planted as augmented partially-replicated (p-rep) designs, obtained genetic and environmental components of variation for a number of yield and tuber components. Heritabilities, trial-to-trial genetic correlations and performance repeatability of clonal selections in p-rep trials and in subsequent fully replicated trial stages were high, and selection was effective for the economically important traits of marketable tuber yield and tuber cooking quality. Simulations using a parameter-based approach, pertaining to the variance components estimated from the p-rep field trials, and the parametric bootstrapping of historic empirical data showed improved rates of genetic gain with p-rep testing over one and two locations compared with testing in fully replicated trials. This potato breeding study suggests that the evaluation and selection of a clonal field crop in fully replicated trials may not be optimal in the early stages of a breeding cycle and that p-rep designs offer a more efficient and practical alternative.     

Association mapping of quality traits in potato (Solanum tuberosum L.)

In this paper, we describe the assessment of linkage disequilibrium and its decay in a collection of potato cultivars. In addition, we report on a simple regression based association mapping approach and its results to quality traits in potato. We selected 221 tetraploid potato cultivars and progenitor lines, representing the global diversity in potato, with emphasis on genetic variation for agro-morphological and quality traits. Phenotypic data for these agro-morphological and quality traits were obtained from recent trials performed by five breeding companies. The collection was genotyped with 250 AFLP^® markers from five primer combinations. The genetic position of a subset of the markers could be inferred from an ultra dense potato map. Decay of linkage disequilibrium was estimated by calculating the squared correlation between pairs of markers using marker band intensities. Marker-trait associations were investigated by fitting single marker regression models for phenotypic traits on marker band intensities with and without correction for population structure. The paper illustrates the potential of association mapping in tetraploid potato, because existing phenotypic data, a modest number of AFLP markers, and a relatively simple statistical analysis, allowed identifying interesting associations.     

马铃薯重要性状QTL定位及3个抗病性状分子标记辅助选育

马铃薯是我国主要粮食作物之一,马铃薯分子育种研究具有重要意义。在二倍体马铃薯中,重要性状控制基因的QTL定位及克隆已经有大量报道。近年,随着同源四倍体分析软件的开发,四倍体马铃薯的遗传图谱构建和QTL定位也取得了突破性进展。分子标记是马铃薯育种的重要辅助手段,可快速准确筛选出多个优良性状。对马铃薯重要农艺性状的QTL定位和克隆,以及3个抗病性状分子标记辅助选育的研究进展进行概述,为加快马铃薯分子育种研究提供参考和实践依据。     

Heterotic effects for yield and tuber solids and type of gene action for five traits in 4x potato families derived from interploid (4x-2x) crosses

The degree of heterosis for total tuber yield (TTY) and total solids (TS) in 4x‐2x crosses was estimated by comparing the performance of 12 families with their respective parents in two locations in Wisconsin (USA). The parental 2x clones were Phureja‐haploid Tuberosum hybrids with 2n‐pollen production by first‐division restitution. The general combining ability (GCA) and specific combining ability (SCA) were estimated for TTY, TS, vine maturity (VM), length of tuber sprout dormancy (LD), and tuber eye depth (ED). Family performance for TTY ranged from 74 to 146% at Hancock (E#1) and from 77 to 287 at Rhinelander (E#2) when compared with that of the 4x parent group. For VM, the families were late maturing, but a few precocious ones were identified. For TS, the families had heterosis of 5.1% over the 4x parent group. The families had slightly higher ED values than the 4x parents, but families with values within the commercial range were identified. The family average for LD (54 days) was closer to the 2x group (51 days) than to the 4x group (88 days). The direction and magnitude of the parent‐family relationships were variable. The 4x parent TTY was correlated with progeny in E#1 but not E#2. The 2x parent VM had correlation with the offspring at E#2 but not at E#1. The type of gene action had a trait‐specific expression. Significant SCA and GCA variances were observed, suggesting that additive as well as non‐additive genetic effects were operating. The 4x‐2x crosses were able to generate heterotic families for TTY and TS in combination with other useful traits. However, no promising results were found for LD because of the apparent dominance of the short‐dormancy phenotype. This result indicates the need of additional selection and breeding efforts for some specific traits when using S. phureja‐derived germplasm.     

Advances in “omics” approaches to tackle drought stress in grain legumes

Grain legumes being affordable sources of proteins, vitamins and essential micronutrients are key to human nutrition worldwide. However, frequent drought episodes present serious threat to grain legume production worldwide. Advances in legume omics in concert with evolving phenotyping and breeding techniques hold great promise to improve drought response of these crops. These resources could underpin prebreeding efforts to expedite discovery and deployment of novel drought tolerance traits into elite backgrounds. Fast-track transfer of traits that confer drought tolerance using marker technologies has been demonstrated in grain legumes like chickpea. However, complex genetic architecture of drought tolerance demands embracing more efficient tools like genomic selection (GS) for accelerated trait improvement. Recent studies on GS for addressing complex traits like drought tolerance have yielded encouraging results in these crops. Recently, speed breeding (SB) protocols have also been optimized for the improvement of long-day/day-neutral grain legumes. Efficacy of SB protocols with regard to complex traits awaits further evidences though. There remains immense scope for integrating SB with GS and gene editing to deliver drought-tolerant cultivars.