Molecular marker-assisted plant breeding on a genome wide scale

Maintaining or increasing agricultural food production and security is a priority in order to meet the needs of a growing population. This challenge is put into further focus by climate change and the need to reduce the environmental footprint of agriculture. There is thus an urgent need to increase the speed of improvement of crop varieties in terms of yield and increased efficiency of use of resources, such as fertiliser and water. Genetic improvement of these traits in crop plants has been achieved by plant breeding on the basis of selection and crossing of phenotypically superior plants. In the last 20 years or so molecular markers have been used in some breeding programmes, but largely on an ad hoc basis for improvement of a few target traits. The advent of more affordable high throughput (next generation) sequencing and genotyping in the last five years has made it possible to make use of molecular markers in a more comprehensive way than hitherto. We refer to genomic selection (GS) which represents a novel way to improve the phenotype of complex agronomic and biological traits governed by many genes each with a small effect. GS is already beginning to transform the breeding of livestock such as cattle and pigs, but has yet to make an impact at a practical level for crop plants. GS is selection based on the collective composition of molecular markers densely covering the entire genome. The proposed collaboration between the Institute of Biological, Environmental and Rural Sciences (IBERS) and the Computer Science Department at Aberystwyth University gives us an opportunity to test GS empirically and theoretically. IBERS is the only university department in the UK with plant breeding programmes, and we will use this unique position by exploiting our perennial ryegrass breeding programme. It is based on repeated cycles of recurrent selection and crossing and is well suited for GS, as we have comprehensive phenotypic data for the current generation and earlier generations of this successful scheme. We will use the current generation of motherplants as a "training population" by genotyping it with over 3000 molecular markers covering the entire genome. The aim is that at least one molecular marker is close to a genomic region influencing the phenotype of interest (quantitative trait locus or QTL). The phenotypic data already available from the breeding programme will be combined with the genotype data to generate complex prediction models using established statistical methods, but also state-of-the-art machine learning techniques developed at the Computer Science Department, for the calculation of a genomic estimated breeding value (GEBV), and to test the performance of the models in the breeding programme. The computational models are then used to calculate the GEBV in a validation population, which is different from the training population, using only genotypic data. The resulting GEBV will be used to select individuals for progeny production based on genotype only. Given a dense coverage of the genome, the combined effect of many QTL for the same trait can be improved measurably by incorporating the effect of all alleles simultaneously. This approach will be particularly advantageous in perennial crops, such as ryegrass and other forages, as the need for lengthy plot trials can be reduced. However, this is not the only benefit of GS. The genomic and statistical resources and models developed here will provide us with a platform for discovery of genes and facilitate the unravelling of the architecture of complex traits of agronomic and biological importance.

The advent of ultra-high throughput DNA sequencing and genotyping generates opportunities to enhance the rate of genetic gain in breeding programmes by combining phenotypic selection with faster molecular breeding approaches. Marker assisted selection has been used largely as an add-on to existing phenotype selection. The objective of this project is to show how genomic selection (GS) in IBERS perennial ryegrass breeding programme can be used to demonstrate that we can reduce the breeding cycle from the current 4 years to 2 years. In GS, many markers are scored across the entire genome in a "training population". This has also been phenotyped, so that models can be developed in which all the markers are used to jointly explain all the genetic variance, and thus make predictions on the breeding value. We will use the 170 and 54 motherplants of the current generation of the two ryegrass breeding populations as training sets as they have been comprehensively phenotyped for a range of traits. We are developing a genotyping platform by next generation sequencing from which we will use a panel of 3072 SNP markers. The genotypic and phenotypic data will be used to generate models for predicting the genomic estimated breeding value (GEBV) with Bayesian regression methods. We will also use state-of-the-art Knowledge Discovery in Databases and machine learning techniques to more efficiently deal with finding predictive relationships between phenotype and genotype. The accuracy of the prediction models will be determined by correlation between GEBV and the true breeding value in half sib progeny from the motherplants, from which the next generation is selected, as well as in historical data and existing varieties. The sequence and genotype resources, and the science underpinning GS that will be developed here has synergy with other plant and animal genetic improvement programmes, and for discovery of genes governing complex traits of agronomic and biological significance.

The main impact will be on plant breeders and geneticists, particularly those concerned with population based genetic improvement programmes. One of the outcomes of this work will be to deliver models and equations for predicting the breeding values of individual genotypes based on genome wide molecular markers. In the particular crop of perennial ryegrass, which is used as the model, selection on the basis of the combined genotype will deliver a time saving of two of the four years of the present phenotypically based breeding cycle. This research thus has great interest for breeding companies and research institutes engaged in genetic improvement of crops.
Increased speed of genetic improvement of forage grasses will also benefit livestock farmers, and have a beneficial impact on food security and consumers.

The project will have wider interests for the academic research community by providing a large number of validated SNP polymorphisms for perennial ryegrass, but more generally will facilitate high throughput molecular marker assisted research to assess the diversity of grass populations and their usefulness for incorporating novel trait characteristics into breeding populations. It will pave the way for more genuinely genome-wide association mapping projects, which have the potential to elucidate individual genes underlying complex traits, such as water soluble carbohydrates and digestibility, to mention a few of high relevance for forage crops. More widely, genomic selection will have a major impact on our ability to unravel the genetic architecture of agronomically and biologically important complex traits, which is key to improving new traits in the longer term.

It will also benefit the research community in terms of contributing models and equations for genomic estimation of breeding value using computational biology and machine learning methods for genomic selection directly relevant for a breeding programme, and thus further demonstrating the beneficial impact of computational biology on genetic and genomics assisted plant research.

The partnership between IBERS and Germinal Holdings Ltd provides us with real opportunities to demonstrate scientific excellence with impact by allowing the potential of GS to be captured for the production of new varieties, and thus showcasing the application of GS in plant breeding.