Genomewide prediction of stability across environments in maize populations

Abstract

Stability models characterize the performance of genotypes across environments. While genomewide prediction is effective for the mean performance of maize ( Zea mays L.) hybrids across environments, its potential for predicting stability has not been reported. Our objective was to assess if genomewide prediction of stability is accurate for yield, moisture, and test weight in maize populations. Stability of performance in seven testcross populations, evaluated in 10–18 environments, was estimated as the regression coefficient from classical Eberhart–Russell analysis. Slope values for each trait in each population had the expected mean of 1.0 and, across populations, ranged from 0.41 to 1.72 for yield (t ha ⁻¹ at 130 g H ₂ O kg ⁻¹ ), 0.65 to 1.27 for moisture (g kg ⁻¹ ), and 0.39 to 1.87 for test weight (g L ⁻¹ ). Across populations, predictive ability for stability ranged from −0.07 to 0.41 for yield, −0.08 to 0.52 for moisture, and −0.18 to 0.62 for test weight. Five different prediction models, including ridge regression and Bayesian models, did not lead to any significant differences ( p  = 0.05) in predictive ability for stability. Predictive ability was always lower for stability than for the across‐environment mean of a genotype. Slopes were also estimated from the predicted performance of each genotype in each environment, but the correlations between observed and predicted slopes remained low. Split‐correlation analysis indicated that the low predictive ability for the slope was due primarily to a low signal‐to‐noise ratio. Our findings indicated that predicting stability in maize is too inconsistent to be routinely useful.