Multi-environment evaluation and genomic prediction of agronomic traits in the southern US rice genepool

The southern US is responsible for 80% of the country’s production of
rice, approximately half of which is exported to other countries.
Understanding genotypic and environmental factors impacting the historical
performance of rice (Oryza sativa L.) is important for directing research
efforts to optimize production of this globally important crop. A set of
429 rice genotypes including globally diverse historical parents and
advanced japonica breeding lines from southern US breeding programs were
phenotyped in 2008 for 8 agronomic traits in Arkansas, Louisiana, and
Mississippi. These were also genotyped using a single-nucleotide
polymorphism set optimized for genomic prediction/selection. Genotypic and
phenotypic data were analyzed via clustering techniques, principal
component analysis, and Finlay-Wilkinson regression. Single trait Genomic
Best Linear Unbiased Prediction, multi-trait genomic prediction (via
mega-scale linear mixed models; MegaLMM), and crop growth modeling
(CERES-Rice in the Decision Support System for Agrotechnology Transfer)
were used to predict/simulate traits on a per-plant basis. We found that
contemporary germplasm from the southern state breeding programs were
highly interrelated and distinct from progenitor indica and temperate
japonica genotypes. Genomic predictive abilities were high and
largely consistent across environments for seed number per panicle, tiller
number, and plant height. Although predictive abilities were lower for
seed weight, that trait was correlated with seed number per panicle (r =
0.919), and predictive ability was higher for both traits in a multi-trait
prediction framework. Furthermore, including data from the two major
genotypic clusters identified herein had no penalty on predictive ability.
The data and analyses presented herein could inform future genomic and
phenotypic investigations and applied breeding in the southern US rice
germplasm pool.

美国南部地区贡献了全美80%的水稻产量，其中约一半出口至其他国家。解析影响水稻（Oryza sativa L.）历史种植表现的基因型与环境因素，对于指导研究工作以优化这一全球重要作物的生产具有关键意义。本研究选取了一套共计429份水稻基因型材料，涵盖全球多样的历史亲本，以及来自美国南部育种项目的高级粳稻（japonica）育种品系；2008年，研究人员于美国阿肯色州、路易斯安那州与密西西比州，对该套材料开展了8个农艺性状的表型鉴定，并采用针对基因组预测/选择优化的单核苷酸多态性（single-nucleotide polymorphism）标记集完成了基因型鉴定。研究人员采用聚类分析、主成分分析（principal component analysis）以及芬莱-威尔金森回归（Finlay-Wilkinson regression）对基因型与表型数据进行了分析。研究人员采用单性状基因组最佳线性无偏预测（Genomic Best Linear Unbiased Prediction）、基于大规模线性混合模型（mega-scale linear mixed models；MegaLMM）的多性状基因组预测，以及作物生长建模（农业技术转移决策支持系统（Decision Support System for Agrotechnology Transfer）中的CERES-Rice模型），对单株水平的性状开展预测与模拟。研究发现，来自美国南部州育种项目的当代种质资源彼此关联性极强，且与原始籼稻（indica）和温带粳稻（temperate japonica）基因型存在显著差异。针对每穗粒数、分蘖数与株高这三个性状，基因组预测精度较高，且在不同环境下整体表现稳定。尽管种子重量的预测精度较低，但该性状与每穗粒数呈显著正相关（相关系数r=0.919）；且在多性状预测框架下，这两个性状的预测精度均有所提升。此外，将本研究鉴定出的两个主要基因型聚类簇的数据纳入分析，并不会对预测精度产生负面影响。本研究提供的数据与分析结果，可为美国南部水稻种质资源库未来的基因组与表型研究以及应用育种工作提供重要参考。