Genomic prediction for stem rust resistance in the southern United states elite oat (Avena sativa L.) germplasm.

Authors: ACHARYA, JANAM - University Of Florida; BABAR, MD - University Of Florida; KHAN, NAEEM - University Of Florida; KUNWAR, SUDIP - University Of Florida; MCBREEN, JORDAN - University Of Florida; ADEWALE, SAMUEL - University Of Florida; Esvelt Klos, Kathy; FURLAN, FLAVIA - Louisiana State University; HARRISON, STEPHEN - Texas A&M University; MELSON, ELLEN - Texas A&M University; HATHCOAT, DANIEL - Texas A&M University; Nandety, Raja Sekhar; Fiedler, Jason

Submitted to: The Plant Genome
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/26/2026
Publication Date: 3/18/2026
Citation: Acharya, J.P., Babar, M.A., Khan, N., Kunwar, S., Mcbreen, J., Adewale, S.A., Esvelt Klos, K.L., Furlan, F., Harrison, S., Melson, E., Hathcoat, D., Nandety, R., Fiedler, J.D. 2026. Genomic prediction for stem rust resistance in the southern United states elite oat (Avena sativa L.) germplasm.. The Plant Genome. https://doi.org/10.3389/fpls.2026.1795871.
DOI: https://doi.org/10.3389/fpls.2026.1795871

Interpretive Summary: Stem rust (SR), caused by the fungus Puccinia graminis f. sp. avenae, is a major disease threatening oat production in the southern United States. Developing durable resistance is critical to protect yield and ensure stable production. Genomic prediction (GP) is a breeding approach that uses genome-wide DNA markers to predict the field response of plants, allowing pre-selection of the best plants within a generation before field data is available. This study evaluated how well various applications of GP would perform when breeding to improve SR resistance. We tested 440 oat lines from the Southern Oat Association Panel, genotyped with two methods. Field trials were conducted across seven environments between 2022 and 2024, assessing disease severity and infection response. Several statistical and machine learning models were compared under three scenarios based on how the genetic data might be used within a breeding program. Models that incorporated simplifying assumptions (the Bayesian models) were best at using genotype data to predict SR resistance in the field, with other types of statistical models performing nearly as well to inconsistently. Prediction of SR resistance using genotype data was most successful when performance of known lines was being predicted for new environments. Both genotyping methods produced similar results. Overall, the results show that SR resistance in oats is largely additive and predictable, highlighting the potential of genomic prediction to accelerate breeding for disease-resistant cultivars and reduce the need for extensive multi-environment testing.

Technical Abstract: Stem rust (SR), caused by Puccinia graminis f. sp. avenae, is a major threat to oat production in the southern United States, necessitating durable resistance to sustain productivity. Genomic prediction (GP) offers a promising approach to accelerate the development of resistant cultivars by leveraging genome-wide marker data to predict breeding values. In this study, we evaluated the accuracy of genomic prediction models for SR resistance using the Southern Oat Association Panel (SOAP), a 440-line multi-institutional panel genotyped with both a 3K SNP array and genotyping-by-sequencing (GBS). Field trials were conducted across seven environments between 2022 and 2024, and disease severity (SV) and infection response (IR) were assessed. Predictive ability (PA) was estimated under three cross-validation (CV) scenarios: CV1 (untested genotypes), CV2 (incomplete field trials), and CV0 (new environments). Across traits and scenarios, Bayesian models (BayesA, BayesB, and Bayesian LASSO) consistently achieved the highest PA, with GBLUP and RRBLUP performing nearly as well, while machine-learning methods (RF and GBR) were less effective. For IR, PA was highest under CV0 (~0.55), moderate under CV2 (~0.50), and lowest under CV1 (~0.45). For SV, PA reached ~0.62 in CV0, ~0.54 in CV2, and ~0.50 in CV1. Differences between the 3K and GBS platforms were minimal, indicating that both genotyping strategies provide sufficient coverage despite contrasting marker distributions and LD patterns. These findings highlight and demonstrate the potential of GS to reduce reliance on large-scale multi-environment phenotyping, enable prediction for new genotypes and environments, and accelerate genetic gain in oat breeding programs.