Metabolic prediction of important agronomic traits in hybrid rice (Oryza sativa L.)

Abstract

Hybrid crops have contributed greatly to improvements in global food and fodder production over the past several decades. Nevertheless, the growing population and changing climate have produced food crises and energy shortages. Breeding new elite hybrid varieties is currently an urgent task, but present breeding procedures are time-consuming and labour-intensive. In this study, parental metabolic information was utilized to predict three polygenic traits in hybrid rice. A complete diallel cross population consisting of eighteen rice inbred lines was constructed, and the hybrids' plant height, heading date and grain yield per plant were predicted using 525 metabolites. Metabolic prediction models were built using the partial least square regression method, with predictive abilities ranging from 0.858 to 0.977 for the hybrid phenotypes, relative heterosis, and specific combining ability. Only slight changes in predictive ability were observed between hybrid populations, and nearly no changes were detected between reciprocal hybrids. The outcomes of prediction of the three highly polygenic traits demonstrated that metabolic prediction was an accurate (high predictive abilities) and efficient (unaffected by population genetic structures) strategy for screening promising superior hybrid rice. Exploitation of this pre-hybridization strategy may contribute to rice production improvement and accelerate breeding programs.

Figure 1. Population structure and hybrid phenotypes in three sub-groups.

(a) Principal component analysis (PCA) of the eighteen rice inbred lines with data for three agronomic traits. Three indica varieties were grouped into the japonica group, and three japonica varieties were grouped into the indica group. Red circles indicate japonica varieties, and blue circles indicate indica varieties. Solid green lines represent the zero values for each component. (b) PCA of the eighteen inbred lines with metabolite profiling data. All 525 analytes were used in the PCA. Two indica varieties were grouped into the japonica group. (c) Dendrogram of the eighteen inbred lines. All eighteen inbred lines were divided into two clear groups. Consistent with the grouping result in (b), only two indica varieties were in the japonica group. The red solid box indicates the japonica group, and the green solid box indicates the indica group. (d) Hybrid phenotypes in three subgroups. The yield per plant (YPP), maturation stage plant height (MSPH) and heading date (HD) were evaluated.

Figure 2. Relationships between PLS latent numbers and R² values.

(a) Sums and differences between parental relative metabolite levels were used as predictive variables for the hybrid phenotypes, relative heterosis and specific combining ability in PLS regression. The R² values varied largely between different traits. (b) Ratios of parental relative metabolite levels were used for hybrid performance prediction. R² values of most traits were highest (above 0.8) at approximately 50 latents. YPP = Yield per plant, MSPH = Maturation stage plant height, HD = Heading date, LPH = Low-parent heterosis, MPH = Mid-parent heterosis, BPH = better-parent heterosis, SCA = Specific combining ability.

Figure 3. Observed and predicted values of YPP, MPH-YPP, BPH-YPP, and SCA-YPP.

(a) Relationships between observed yield per plant and predicted yield per plant. Predicted values were calculated with the equations based on variable coefficients in the PLS regression results. The horizontal axis represents the observed values, and the vertical axis represents the predicted values. The green solid line represents the total fit line, and the black dotted line is y = x. (b–d) Relationships between the observed relative mid-parent heterosis, better-parent heterosis, specific combining ability of YPP and the corresponding predicted values. YPP = Yield per plant, LPH = Low-parent heterosis, MPH = Mid-parent heterosis, BPH = better-parent heterosis, SCA = Specific combining ability.

Figure 4. Observed and predicted YPP values in three subgroups and two reciprocal groups.

(a–c) Relationships between observed yield per plant and predicted values in the i, j and ij three subgroups. The entire hybrid population was divided into three subgroups based on whether a parental line was indica or japonica (Fig. 1b). The predictive abilities in different population structures showed only slight changes. (d–e) Relationships between observed yield per plant and predicted yield per plant in reciprocal hybrids. The whole population was divided into two groups of reciprocal hybrids to test cytoplasmic effects on predictive ability. The results demonstrated that different cytoplasms only weakly influenced predictive ability. YPP = Yield per plant.