Multi-trait multi-environment genomic prediction of agronomic traits in advanced breeding lines of winter wheat

Authors: GILL, HARSIMARDEEP - South Dakota State University; HALDER, JYOTIRMOY - South Dakota State University; ZHANG, JINFENG - South Dakota State University; BRAR, NAVREET - South Dakota State University; HALL, CODY - South Dakota State University; RAI, TEERATH - South Dakota State University; Bernardo, Amy; St Amand, Paul; Bai, Guihua; OLSON, ERIC - Michigan State University; ALI, SHAUKAT - South Dakota State University; TURNIPSEED, BRENT - South Dakota State University; SEHGAL, SUNISH - South Dakota State University

Submitted to: Frontiers in Plant Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/12/2021
Publication Date: 8/18/2021
Citation: Gill, H., Halder, J., Zhang, J., Brar, N., Hall, C., Rai, T., Bernardo, A.E., St Amand, P.C., Bai, G., Olson, E., Ali, S., Turnipseed, B., Sehgal, S. 2021. Multi-trait multi-environment genomic prediction of agronomic traits in advanced breeding lines of winter wheat. Frontiers in Plant Science. https://doi.org/10.3389/fpls.2021.709545.
DOI: https://doi.org/10.3389/fpls.2021.709545

Interpretive Summary: Genomic selection is a new crop breeding method that uses statistical models and a large number of genome-wide DNA markers to predict performance of breeding lines without the need for extensive field testing. Genomic selection models can be based on single traits or on multiple correlated traits. We constructed models using 314 advanced US hard winter wheat lines that were evaluated at five locations over two years. The genomic selection models were compared for predicting five agronomic traits at the ten individual environments. Overall, multi-trait models performed significantly better than the single-trait models. Implementing the multi-trait genomic selection models can improve selection efficiency in wheat breeding programs.

Technical Abstract: Genomic selection (GS) is a promising approach for accelerating the genetic gain of complex traits in wheat breeding. However, increasing the prediction accuracy (PA) of GS remains a challenge in the successful implementation of this approach. Multivariate approaches can leverage simultaneous evaluation of several traits under multiple environments by exploring correlations to improve GS performance in breeding programs. Here, we compared GS models to predict single and multiple agronomic traits using 314 advanced breeding lines of winter wheat that were evaluated at five locations over two years (2019 and 2020). The multi-trait (MT) model was evaluated with two cross-validation schemes representing different breeding scenarios (CV1, prediction of completely unphenotyped lines; and CV2, prediction of lines partially phenotyped for correlated traits). We also evaluated the Bayesian multi-trait multi-environment (MTME) model that integrates the analysis of multi-traits evaluated under multi-environments. The GS models were compared for predicting five agronomic traits at ten individual site-years. The MT-CV2 model outperformed all other models for predicting grain yield, with improved PA ranging from 25% to 100% over the single-trait model. The MTME model performed better for all traits except grain yield, with improvement over the ST-CV1 reaching up to 100%, 43%, 71%, and 86% for grain protein content, test weight, plant height, and days to heading, respectively. Overall, both MT-CV2 and MTME models outstripped the single-trait model to predict various agronomic traits in most environments. The increase in PA using MT-CV2 was attributed to a moderately-high degree of trait-correlations in different environments. Similarly, the correlation between environments within a season was crucial for prediction improvement of the MTME model. Our results demonstrate that the multivariate GS models hold great potential in implementing GS in breeding programs.