Assessing the potential of genomic selection to improve yield and persistence in white clover : a thesis presented in the partial fulfilment of the requirements for the degree of Doctor of Philosophy (PhD) in Plant Biology at Massey University, Manawatu, New Zealand

Abstract

White clover (𝘛𝘳𝘪𝘧𝘰𝘭𝘪𝘶𝘮 𝘳𝘦𝘱𝘦𝘯𝘴 𝘓.) is an economically important forage legume in temperate pastures, providing quality fodder and plant-available nitrogen. However, its potential has not been fully exploited due to unpredictable herbage yield and poor vegetative persistence in pasture. Identification of genotypes that combine traits essential for yield and vegetative persistence, like dry matter yield and stolon density, are key objectives in breeding programmes. Long breeding cycles, high genome complexity and difficult-to-phenotype traits, usually assessed at late growth stages, are major constraints to conventional phenotypic selection in white clover breeding. In cultivar development programmes, elite individuals must be accurately identified and selected before crossing to generate superior progeny. Genomic selection is becoming a preferred method for increasing the rate of genetic gain by enabling early identification and selection of superior individuals, based on their genomic estimated breeding values (GEBVs), which can be generated without the need for phenotyping. Genomic selection is usually performed using a statistical model developed using genotypic and phenotypic information derived from a training population. In forage breeding, as parental breeding values are estimated by progeny testing, phenotypic data used in genomic prediction models is obtained from half-sib progeny. Recent single nucleotide polymorphism (SNP) genotyping methods like genotyping by sequencing (GBS) which generate a large volume of SNP marker information at low cost, have made genomic selection possible for species such as white clover. The main objective of this thesis was to explore the potential of genomic selection to improve important traits in white clover breeding.

A training population of 274 white clover parents were genotyped using GBS to provide genotype information. These genotyped maternal parents were randomly polycrossed under isolation to generate 274 HS families from which 200 HS were selected for phenotyping. The HS families were established in replicated, multi-location mixed sward field trials in 2016 at Aorangi and Ruakura New Zealand, under dairy cattle grazing. Variance components and quantitative genetic parameters were estimated from the HS progeny families via Residual Maximum Likelihood (REML) analyses for traits dry matter yield, growth score, leaf size, stolon number, stolon branching and Hydrocyanic acid (HCN) production. There was significant (P < 0.05) additive genetic variation among HS families for all measured traits. Year, season and location effects were also significant. Family mean narrow-sense heritability for the traits ranged from low (0.13) to high (0.82). There was a low but positive correlation (0.24) between DM yield and stolon number. Results from cluster analysis identified several HS families with high DM yield and stolon density.

Predictive ability assessed by Monte-Carlo cross validation, ranged from -0.17 to 0.44 for different traits. Predictive ability for dry matter (DM) yield from data merged across years and environments was 0.3, while stolon density traits, stolon number and branches had lower predictive abilities ranging from -0.17 to 0.21. The highest predictive ability, 0.44, was obtained for leaf size, a genetically less complex trait than the yield-associated traits.

The performance of different genomic prediction models, Genomic BLUP (GBLUP), KGD-GBLUP, BayesCπ, and Reproducing Kernel Hilbert Spaces (RKHS) were compared. While no significant difference in predictive ability among models was detected, KGD-GLUP, a very computationally efficient model, tended to generate the highest predictive abilities on average. There was no decrease in predictive ability when the number of individuals in the training population and SNP markers were reduced from 200 and 110,000 to 80 and 5,500, respectively. Multi-trait genomic selection in which primary and secondary traits are incorporated into the model, increased predictive ability only when the information of a highly correlated secondary trait was present in both the training and test populations.

Using simulation, it was demonstrated that an integrated strategy using conventional phenotypic selection to select among families and genomic selection to select within families, termed AFp-WFgs, delivered up to two-fold genetic gain for DM yield over conventional phenotypic selection among families alone by enabling access to the ¾ additive variation residing within HS families. The cost efficiency of implementing genomic selection was also investigated and showed AFp-WFgs was more cost-efficient than among family phenotypic selection under high selection pressures.

Finally, to empirically validate the obtained predictive ability, a divergent selection was conducted for a simple trait, HCN, by selecting individuals based on their GEBVs. Conventional among HS family selection, progeny test selection and AFp-WFgs were compared in terms of response to selection, genetic gain and accuracy of selection. Despite the low predictive ability of 0.22 obtained for HCN, results showed AFp-WFgs to be similar to progeny test selection and superior to phenotypic selection in terms of response to selection and genetic gain. In terms of accuracy, AFp-WFgs was the more accurate selection method, successfully eliminating individuals with high or low HCN production in the low and high HCN divergent groups, respectively.

Our results indicate, for the first time, an integrated phenotypic and genomic selection approach to be superior to conventional phenotypic selection at increasing genetic gain for a simple trait in white clover. This demonstrates the potential of genomic selection to be used in enhancing white clover breeding programmes for quantitative trait improvement.