NIRS calibrations for key forage quality traits impacting animal performance (including organic matter digestibility, neutral detergent fiber, and crude protein) were developed and used to phenotype a large training population that had been sampled for forage quality analysis across multiple years and seasons (over 14,000 samples). In building calibrations, a number of spectral and mathematical pretreatments were evaluated and optimal treatments selected; resulting in development of calibrations with strong predictive performance. A Shiny application created to assist calibration development has been published from this work. These data were first used to determine genotypic and environmental variance components for forage quality traits in this material, and to identify spectral bands strongly associated with quality parameters. The study looked at two complementary quality traits: one determining the digestible portion of the sample, and one related to the estimation of the un-digestible portion. It was also found that time of cut, days-to-heading and precipitation measured over 28 days were key parameters for ash and crude protein; whereas sum of mean temperature measured over 14 and 28 days were the most important parameters impacting digestibility. NIRS calibrations can now be exploited to phenotype forage material during routine breeding, and the phenotyping of the training population can be exploited to identify families, cultivars, and accessions with excellent forage quality for use as germplasm in new cycles of breeding.
The phenotype data above was combined with existing genotype data on the training population to develop genomic prediction equations for forage quality traits. Predictive modeling was carried out using a number of different approaches; these included approaches based on using the DNA data directly and approaches using the genomic relationship matrix derived from the DNA data. Model performance was evaluated using two methods of cross-validation, (i) setting aside random sets of individuals for testing or (ii) leaving out complete families/cultivars/accessions from the training data for testing. Results of the latter highlighted the importance of close relationship between training data and selection candidates for effective implementation of genomic selection in forage breeding. Encouragingly, good predictive accuracies were achieved for forage quality traits, highlighting their potential to use GS for indirect selection. To further validate these predictive models in subsequent generations of the training population, samples were collected from a field trial of F2 families derived from crossing plants of the training population. These samples were collected from two trials (over 1,200 samples) and scanned using NIRS to determine forage quality parameters using developed calibrations. F2 families were also genotyped and a common marker set used to validate the predictive models developed on the training population. The encouraging results demonstrated the potential of using genomic selection to improve forage quality in perennial ryegrass breeding, particularly given that GS will be implemented on seedlings from F2 families. These models are now being exploited to quantify the added value of among-and-within-family genomic selection through an empirical evaluation. A range of synthetic cultivars will be developed using positive genomic selection, negative genomic selection, and random selection, and compared using field evaluations to quantify genetic gains associated with among-and-within-family genomic selection.
A genome wide association study was also carried out using the available data to identify DNA polymorphisms associated with forage quality traits, and a number of significant marker-trait associations were found. In addition to the GWAS, a study was carried out to evaluate the use of genetic algorithms to identify important variables contributing to predictive accuracy of genomic selection and the work disseminated at an international conference.
