Rice (Oryza sativa L.) is the world's most important staple food crop, especially in Asia. As a semi-aquatic crop species, water-scarcity and increasing severity of water-deficit stress owing to climate change, are a major threat to sustaining irrigated rice production. Improving the rice adaptation to water-deficit is, therefore, a primary breeding target. The main goal of this dissertation is to study the morphological, anatomical, physiological and genetic basis for responses of a rice plant to water-deficit stress.
To give leads into how water-deficit tolerant rice should behave, a comparative study were conducted, whereby representative rice genotypes was compared at the same moisture stress during the vegetative stage with genotypes of wheat, a dryland cereal wheat (Triticum aestivum L.) known to be more tolerant to water-deficit than rice. Under-water-deficit, rice genotypes (IR64 & Apo) developed thinner roots allowing rapid water-acquisition, whereas wheat followed a water-conserving strategy through developing thicker leaves and roots, and moderate tillering. Root anatomy such as root diameter, xylem and stele diameter and xylem number were more plastic in wheat than in rice under-water-deficit.
The methodology and findings from those representative genotypes were then projected to a diverse panel of nearly 300 rice genotypes. Such a panel was previously constructed by the International Rice Research Institute as a potential means of discovery of novel beneficial alleles for diverse phenotypic traits and their plasticity, with 46K high-quality single nucleotide polymorphisms (SNPs). A genome-wide association study (GWAS) was undertaken to identify the genomic regions regulating the morphological, physiological and root anatomical traits in rice, based on a large-scale greenhouse phenotyping of these traits. The genetic basis of these traits was different in control and water-deficit stress (strong quantitative trait loci [QTL] × environment interaction), in line with novel loci detected for the plasticity of traits. Key a priori candidate genes near to these genetic loci were also identified.
Rice grain yield is strongly affected by water-deficit stress coinciding with sensitive reproductive stage. Strong genotypic variability for grain yield as well as yield components and related traits were observed in the same rice indica diversity panel, under control and reproductive stage water-deficit stress in field conditions across two years. The GWAS analysis identified the core loci of rice genome governing the grain yield and related traits. Most of the genomic loci were specific to treatment and year, indicating strong QTL × environment interactions.
To enable GWAS findings to be used for better designing of genotypes by breeding, an existing process-based crop model GECROS was used in a case study, where grain yield of the same indica diversity panel (267 rice genotypes) from the control treatment in one season was dissected into eight physiological parameters. Some parameters had a stronger effect on grain yield than other parameters. Using these parameters, the model showed the ability to predict the genotypic variation of rice diversity panel for grain yield under different field conditions. Further, the GWAS analysis was extended to model-input parameters on randomly chosen 213 genotypes as a training dataset. The SNP-based estimates of parameter values calculated from the additive allelic effect of the loci were used as input to the crop model GECROS. Although the SNP-based modelling approach demonstrated the ability to predict the genotypic variation in training datasets under different environments, the prediction accuracy was lower in the remaining 54 genotypes used as a testing dataset. In addition, the prediction accuracy of grain yield was also lower using either parameter or SNP-based GECROS model in completely new season. However, the model-based sensitivity analysis effectively identified the different SNPs between control and water-deficit environments. Virtual ideotypes designed based on pyramiding the SNPs identified by modelling had a higher yield than those based on SNPs for yield per se.