QTL mapping and genomic prediction of seedling root architectural traits in a rice RIL population.

Authors: Sharma, Santosh; Edwards, Jeremy; Gealy, David; Pinson, Shannon

Submitted to: Rice Technical Working Group Meeting Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 12/4/2019
Publication Date: 1/6/2021
Citation: Sharma, S., Edwards, J., Gealy, D.R., Pinson, S.R. 2021. QTL mapping and genomic prediction of seedling root architectural traits in a rice RIL population. Proceedings of 38th Rice Technical Working Group Meeting, February 24-27, 2020, Orange Beach, Alabama. p 83-84. Electronic Publication.

Interpretive Summary:

Technical Abstract: Rice root system architecture (RSA) is crucial for resource acquisition in both stress and non-stress environments. However, these traits are very difficult to phenotype. High-throughput automatic root image analysis systems can help expedite the genetic study of RSA. Greater understanding of the genetics controlling RSA and the interactions with important agronomic traits could lead to new marker assisted selection (MAS) and genomic selection (GS) tools for breeders. These tools could be especially important for difficult to phenotype traits like roots. The objectives of this study were to (1) genotype and phenotype rice seedlings for RSA in a population derived from parents differing in root architecture, (2) to discover quantitative trait loci (QTL) for RSA traits, (3) to evaluate the accuracy of genomic prediction methods for RSA traits, and (4) to model the network of interactions among the RSA traits, RSA QTLs, and their interactions with above-ground agronomic traits. A recombinant inbred line (RIL) population (PI312777 x Katy) was genotyped with a 7K SNP array and phenotyped for seedling RSA and agronomic traits. QTL mapping and genomic prediction were used to study variance in RSA traits evaluated on agar plates with WinRHIZO, along with their associations with agronomic traits measured in the field. The filtering of SNPs for non-redundancy and QTL mapping were performed with the inclusive composite interval mapping (ICIM) method implemented by the ICIMAPPING program. The genomic prediction methods evaluated included rrBLUP, GBLUP and BRR as implemented in the R packages rrBLUP, Sommer and BGLR respectively. QTLs for multiple quantitative traits and trait interactions were modeled using the “Bnlearn” Bayesian network analysis package in R and evaluated with k-fold validation. Two QTLs for percent root hair surface area (of the total root surface area) were detected on chromosomes 11 and 5 explaining 33.9% of the variance. Two QTLs for root surface area were detected on chromosome 11 explaining 29.6% of the variance. One QTL for percent root hair length (of the total root length) on chromosome 5 and one QTL for number of root tips on chromosome 12 were detected explaining 34.3 and 14.9% of the variance respectively. The genomic prediction accuracies for percent root hair surface area were 58.6, 58.3 and 58.2% and root surface area was 32.3, 27.5 and 34.1% by rrBLUP, GBLUP and BRR method respectively. The prediction accuracies for percent root hair length (of the total root length) were 59.2, 57.9 and 54.1%; and 47.3, 37.0, and 55.2% for number of root tips. Among the root and agronomic traits, we found that cross-validated genomic prediction methods had greater accuracy (explained higher proportion of genetic variance) than single or multi-QTL models. However, the identified QTL regions can be used for MAS with a small number of markers, and the QTLs may be further investigated for fine mapping and gene cloning. Bayesian network analysis revealed multi-trait QTLs and the relationships between traits. Between the RSA traits, root length, surface area, volume and number of root tips showed significant relationships as well as root hair length, root hair surface area, and number of root hair tips. These relationships were expected because of the mathematical dependence in calculations of the RSA traits. Across RSA and agronomic traits there were significant relationships between root hair length and yield, between root tips and heading date, plant height and number of root hair tips. Some of the same RSA QTLs found by single trait ICIM were found by the Bayesian network analysis, including QTLs for percent root hair surface area on chromosome 11. The Bayesian network analysis also revealed QTLs for RSA traits that were not detected by single trait ICIM. These include a QTL on chromosome 1 for root length proximal to the O