Enhancing alfalfa breeding through genomic prediction with exotic germplasm resources

Authors: CHEN, SHUFEN - Cornell University; LIN, MENG - University Of Florida; Tilhou, Neal; BASNET, BHOJA - Cornell University; ZHAO, DONGYAN - University Of Florida; BEIL, CRAIG - University Of Florida; Riday, Heathcliffe; SHEEHAN, MOIRA - University Of Florida

Submitted to: Plant and Animal Genome Conference Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: 12/25/2025
Publication Date: 1/12/2026
Citation: Chen, S., Lin, M., Tilhou, N.W., Basnet, B., Zhao, D., Beil, C., Riday, H., Sheehan, M. 2026. Enhancing alfalfa breeding through genomic prediction with exotic germplasm resources. Plant and Animal Genome Conference Proceedings. 1.

Interpretive Summary: Alfalfa is an important crop in the United States. It is challenging to develop improved alfalfa varieties for farmers. Breeding progress could be improved by collecting new genetic diversity from alfalfa's native range in Eurasia, but including this germplasm into modern breeding programs is slow. Genomic prediction models may be able to accelerate this process. This study evaluated the performance of genomic prediction models and concluded that models must focus on including both a large number of training individuals and a diverse mix of training individuals. This research is an important first step in harnessing new alfalfa germplasm.

Technical Abstract: Alfalfa (Medicago sativa L.) is the most important forage legume globally, valued for its high biomass yield and nutritional value in animal feeding. To establish effective genomic prediction workflows for alfalfa using exotic germplasm, a set of 780 alfalfa samples collected from different geographic origins (CASIA, EURO, OTTM, and SIYBR) and U.S. breeding materials (CHECK) were evaluated in a multiple-year field trial for growth habit (GH), plant height (HGT), and plant vigor (VIG). Analysis of the alfalfa 3K DArTag panel genotyping results revealed distinct linkage disequilibrium patterns across these five subgroups of germplasm. Genomic prediction across the entire population outperformed predictions within individual subgroups, suggesting that combining multiple subgroups enhanced genetic diversity and predictive accuracy. To examine the role of training set composition using multiple subgroups of exotic germplasms, we evaluated schemes incorporating varying proportions (10–50%) of the predictive subgroup in the training set and compared them with the scheme lacking shared subgroup information between training and testing sets. Increasing the proportion of the predictive subgroup gradually improved predictive performance, highlighting the value of including genetically related lines. Building on this, further optimization was achieved by systematically varying the proportion of samples from all five subgroups in the training set (10–90%), showing that predictive ability increased with training-set size. Overall, these results indicate that successful genomic selection in alfalfa requires not only expanding training-set size but also strategically incorporating exotic germplasm and evaluating newly introduced base populations to maximize prediction accuracy.