Assessing the agronomic potential of sorghum B-lines using genomic prediction

Authors: KENT, MITCHELL - Texas A&M University; FONSECA, JALES - Texas A&M University; KLEIN, PATRICIA - Texas A&M University; Klein, Robert; Hayes, Chad; ROONEY, WILLIAM - Texas A&M University

Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/11/2023
Publication Date: 10/9/2023
Citation: Kent, M.A., Fonseca, J.M., Klein, P.E., Klein, R.R., Hayes, C.M., Rooney, W.L. 2023. Assessing the agronomic potential of sorghum B-lines using genomic prediction. Crop Science. 63(6):3367-3381. https://doi.org/10.1002/csc2.21107.
DOI: https://doi.org/10.1002/csc2.21107

Interpretive Summary: The yield potential in grain sorghum hybrids has increased at a slower rate than other cereal crops including its close relative maize. While there are many reasons for this lag, increasing hybrid performance through genomic selection has the potential to accelerate the rate of genetic gain in sorghum to levels that parallel gains in hybrid maize. To address this issue, we implemented a program to determine whether we can predict the yield potential of seed parents using genomic prediction models. This work provides the necessary knowledge to breeders who work to exploit genomic technologies in improving grain yield of hybrid cereal crops including sorghum.

Technical Abstract: In hybrid sorghum breeding, basic agronomic traits such as days to flowering and plant height of sorghum seed parents (B-lines) must be within a specific range for hybrid seed production. Because many sorghum programs select new B-lines outside of commercial seed production environments, the purpose of this study was to determine if genomic prediction is effective to eliminate new lines that do not fit the range for these traits. Two B-line RIL populations were evaluated across several environments for days to mid anthesis (DY), plant height (PH), panicle length (PL) and grain yield (GY). Across environments and populations, average prediction accuracies were between 0.47 and 0.61 for DY, 0.24 and 0.60 for PH, 0.16 and 0.37 for GY, and 0.37 and 0.57 for PL. The effect of training set size was assessed by subsampling various amounts of data into the training set, ranging from 5% to 65%. Prediction accuracies generally improved as the proportion of the total data in the training set increased for both inter- and intra-population predictions, but a relatively small portion (15%) of the total data still produced modest prediction accuracies. The effect of the marker density on genomic prediction accuracies was assessed by subsampling various numbers of SNPs. Marker density as low as 500 markers had a minimal effect on the mean or the range of prediction accuracies. The results of this study indicated that genomic prediction can be a useful and cost-effective tool that sorghum breeding programs should incorporate into their pipeline.