Genetic association analysis of indica rice yield and related traits in irrigated ecosystems

Rice is the dominant food source for more than half of the world's population. Irrigated rice predominates in global rice production as well as in most rice producing countries. Further improvement of grain yield (GY) by combining well-proven conventional breeding methods with new techniques offered by modern molecular biology and genomics is urgently needed to feed the growing world population. The studies reported in this thesis are part of the effort led by IRRI to break the yield barrier for irrigated rice. This research provides essential phenotypic and genetic information directly relevant to future breeding for irrigated ecosystems. The specific objectives are: i) to gain a better understanding of the importance of genotype-environment interaction (GEI) on GY in irrigated rice ecosystems; ii) to test the usefulness of 39 fine-mapped or cloned genes/QTLs for GY and yield related traits in a breeding population through association analysis and iii) to identify new markers associated with GY and related traits through a genome-wide association study (GWAS). The studies were conducted using a collection of 392 genotypes including released cultivars and advanced lines from several large breeding programs worldwide. These are being used as parental lines in IRRI's recurrent selection and variety development programs. Field trials were conducted in eight environments including Jiangxi (JX) and Sichuan (SC) in China, and six season (2) and nitrogen rate (3) combinations at IRRI headquarters (Philippines). The two seasons were the dry season (DS) and wet season (WS) of 2012. For the DS the three nitrogen rates were no nitrogen, 90 kg ha-1 and 180 kg ha-1, designed as DS1, DS2 and DS3. For the WS the three nitrogen rates were no nitrogen, 45 kg ha-1 and 90 kg ha-1, designed as WS1, WS2 and WS3. GY and the following 10 traits were measured, including grain number per panicle (GN), panicle number per plant (PN), thousand grain weight (TGW), days to flowering (DTF), primary branch per panicle (PB), plant height (PH), secondary branch per panicle (SB), Spikelet number per panicle (SN), seed setting rate (SR) and tiller number (TN). All the 11 traits were tested in DS1, DS3, WS1, WS2 and WS3, while selected traits were tested in DS2, JX and SC. A wide range of variations across genotypes and environments were observed for all traits. Genotype, environment and GEI all significantly affected GY and yield related traits. GEI was more important than the genotypic main effect for GY, SR and PN but less important for other traits. For GY, the genotype-by-season interaction and genotype-by-season-by-nitrogen interaction were more important than the genotype-by-nitrogen interaction. The genotypes were clustered into 10 groups using an agglomerative hierarchical clustering procedure. The eight environments were grouped into three groups using the biplot of the additive main effects and multiplicative interaction (AMMI) analysis. The three environments (nitrogen rates) in the WS and SC were grouped together (E1) and three environments (nitrogen rate) in the DS formed another group (E2). JX alone was the third group (E3). JX was relatively closer to the IRRI DS environments in the biplot. It indicated that IRRI breeding lines with stable and good performance in the WS could be used in SC (directly as varieties or as parental lines in breeding) and that selection is better done in the DS in IRRI for use in JX, China. Great attention should be paid to the relevance of performance at IRRI to their target production environments when IRRI breeding lines are introduced. Breeding for DS and WS separately at IRRI was recommended to exploit the repeatable GEI caused by seasonal variation. To test the usefulness of fine-mapped or cloned genes/QTLs for GY and yield related traits, 46 molecular markers tightly linked to the chosen 39 genes/QTLs were used to genotype 360 of the 392 lines. Population structure analysed with 53 SSR markers evenly distributed on all chromosomes using STRUCTURE program indicated that the whole population could be divided into two subpopulations of 205 and 155 lines. A mixed linear model incorporating genetic relatedness between genotypes was chosen by comparing four commonly used statistical models. The selected model was used to conduct association analysis for all the tested traits in each of the eight testing environments and the average environment defined as the average across the testing environments. Analyses were separately carried out for the whole population and the two subpopulations. All the 39 target genes/QTLs were associated with two or more measured traits including traits not previously reported. GW6 and Gn1a were associated with nine and eight traits, respectively. Ghd7, qSPP7, SCM2 and SPP1 were associated with seven traits, GIF1 and Ltn were with six traits, GS3, GW2, gw3.1, htd1, Nop(t), qGY2-1 and qPH6-1 with five traits, D10, d27, DEP2, DWL1, Gnp4, Gw1-1, GW3, gw5, MOC1, PAP2, qGL7, qGL7-2 and qGN4-1 with four traits, D88, Ghd8, GS5, Gw1-2, IPA1, qSH3 and RPH with three traits and ep3, gw8.1, gw9.1 and qPDS3 with two traits. A total of 16 genes/QTLs were found to be associated with GY. GS3, GW1-1 and d27 were associated with GY in two testing environments and the others were only in one. For the three yield component traits GN, PN and TGW, there were 16, six and 10 genes/QTLs identified to be associated with in one or more environments. Eleven genes/QTLs were associated with SN and SR, respectively. There were 29 and six genes/QTLs detected to be associated with PB and SB, respectively. TN had the least number of significant genes/QTLs associated, which were five. All the 39 genes/QTLs were associated with PH in one testing environment or the average environment. There were 25 genes/QTLs found to be associated with DTF in one or more environments. Significant gene- by-environment interaction was present for all the studied genes/QTLs. However, GY could not be well predicted using the markers significantly associated with measured traits or all the target markers based on stepwise multiple linear regression analysis. The adjusted coefficient of determination (R2) ranged from 0.024 to 0.191 for the final selected models considering the associated markers only and from 0.039 to 0.261 for the final selected models considering all the target markers. Nevertheless the known genes might be explicitly utilized in developing more efficient selection criteria for enhancing selection accuracy. To identify new markers associated with GY and related traits, 327 of the 392 lines were genotyped with SNPs using the genotyping-by-sequencing method. Model based population structure analysis was conducted with a subset of 1072 evenly distributed SNPs. The results indicated that the likely number of subpopulations was two, with subpopulation 1 consisting of 234 lines and subpopulation 2 consisting of 93 lines. There were 56 common lines between the two smaller subpopulations derived from the population structure analysis results using SSR and SNP. Based on comparison of multiple models for selected trait and environment combinations, the PK model implemented in TASSEL software using principle components to correct population structure and the relative kinship matrix to adjust the unequal familial relatedness between the individuals was selected to conduct the GWAS for all the traits in all the testing environments. A total of 452 marker-trait associations that were delineated into 43 QTLs were identified for all traits but PB, SB and SR with 39 QTLs being not reported before. Three QTLs were identified for GY on chromosome 6, 9 and 12 but only in DS2. The numbers of QTLs identified for PN, GN and TGW were 26, four and two. There were nine and four QTLs detected for DTF and TN, respectively. Two QTLs were identified for PH and SN, respectively. Most of the detected QTLs were found in only one environment. One of the QTL for DTF on chromosome 3 was identified in multiple environments and corresponds to Hd9 reported in previous studies. Two of the QTLs for PN on chromosome 1 were in the regions of previously fine-mapped QTLs, Gw1-1 and Gw1-2, for TGW. The effects of newly identified QTLs were relatively small with the highest percentage of variance explained by a single QTL being 9.6%. Gene-by-environment interaction, pleiotropy and small effects of the well characterized genes/QTLs and newly detected QTLs imply that selection accuracy using the identified genes/QTLs is low. Improving yield and related quantitative traits through marker-assisted selection remains a big challenge. Recently developed genomic selection that utilizes markers in linkage disequilibrium with all genes affecting trait of interest and captures interactions between genes should be exploited.