Increased Prediction Accuracy Using Combined Genomic Information and Physiological Traits in A Soft Wheat Panel Evaluated in Multi-Environments

Jia Guo,Md Ali Babar,Sumit Pradhan,Jahangir Khan,J Paul Murphy,Jordan Mcbreen,Dipendra Shahi,Guihua Bai

doi:10.1038/s41598-020-63919-3

Abstract

An integration of field-based phenotypic and genomic data can potentially increase the genetic gain in wheat breeding for complex traits such as grain and biomass yield. To validate this hypothesis in empirical field experiments, we compared the prediction accuracy between multi-kernel physiological and genomic best linear unbiased prediction (BLUP) model to a single-kernel physiological or genomic BLUP model for grain yield (GY) using a soft wheat population that was evaluated in four environments. The physiological data including canopy temperature (CT), SPAD chlorophyll content (SPAD), membrane thermostability (MT), rate of senescence (RS), stay green trait (SGT), and NDVI values were collected at four environments (2016, 2017, and 2018 at Citra, FL; 2017 at Quincy, FL). Using a genotyping-by-sequencing (GBS) approach, a total of 19,353 SNPs were generated and used to estimate prediction model accuracy. Prediction accuracies of grain yield evaluated in four environments improved when physiological traits and/or interaction effects (genotype × environment or physiology × environment) were included in the model compared to models with only genomic data. The proposed multi-kernel models that combined physiological and genomic data showed 35 to 169% increase in prediction accuracy compared to models with only genomic data included when heading date was used as a covariate. In general, higher response to selection was captured by the model combing effects of physiological and genotype × environment interaction compared to other models. The results of this study support the integration of field-based physiological data into GY prediction to improve genetic gain from selection in soft wheat under a multi-environment context.

Highlights

Citra 2015-2016 Citra 2016-2017 Citra 2017-2018 Quincy 2016-2017 Citra Normal Quincy Normal grain yield and biomass partitioning traits that are affected by large numbers of small-effect genes8
Aguate et al.12 indicated that integrating all hyperspectral wavelengths using ordinary least squares, partial least squares, and Bayesian shrinkage resulted in higher prediction accuracy than using individual vegetation indices in maize
Applying high throughput phenotyping on SPAD, canopy temperature (CT), membrane thermostability (MT), and normalized difference vegetation index (NDVI) during milk-grain stages could potentially accelerate selection and advancing germplasm in wheat

Summary

Introduction

Citra 2015-2016 Citra 2016-2017 Citra 2017-2018 Quincy 2016-2017 Citra Normal Quincy Normal grain yield and biomass partitioning traits that are affected by large numbers of small-effect genes. Krause et al. used genomic marker-, pedigree-, and hyperspectral reflectance-derived relationship matrices to construct genomic-enabled BLUP models to evaluate the genetic main effects (G) and GE interactions across environments in a wheat breeding program and showed the highest prediction accuracies when combining marker/pedigree information with hyperspectral reflectance phenotypes. Pérez et al. applied a Bayesian-based prediction model utilizing both molecular markers and pedigree information and extended it to a multi-kernel prediction model suitable for combining multiple omic data. Pérez et al. applied a Bayesian-based prediction model utilizing both molecular markers and pedigree information and extended it to a multi-kernel prediction model suitable for combining multiple omic data28 This approach was proved to increase prediction accuracies in maize and wheat. The objectives of this study were to: 1) propose models using genomic and field-based physiological data to predict the grain yield in a soft facultative wheat panel, 2) compare the prediction accuracies of the model that combined field-based physiological traits with genomic data under a multi-environment context to the model that was built on either physiological traits or genomic data, and 3) rank the importance of contribution by different physiological traits to grain yield

Objectives

Methods

Results

Discussion

Conclusion