Introduction

Rice, a staple food crop feeding majority of the world population, is cultivated in 162.05 million hectares with the production of 755.43 million tonnes (FAOSTAT, 2019). India contributes closer to 20% of the world’s rice production, and Indian rice production should increase by at least 50% by 2050 to meet the requirements of increasing population. But, any further increase in rice production is challenged by yield plateau, declining natural resources and increased occurrence in the frequency of biotic/abiotic stresses (Ray et al., 2013; IndiaStat, 2020). This warrants development of high yielding and climate resilient rice varieties through systematic exploitation of rice genetic diversity and effectively introgress the desirable alleles linked to grain yield and stress tolerance traits. Grain yield in rice is a complex traits contributed by several sub-components. There are several yield genes namely Gn1a, DEP1, DEP2, NAL1/SPIKE, etc., have been reported to control panicle related traits in rice and thereby controlling grain yield potential (Li et al., 2021). But, many of them have been discovered in japonica rice and their effects are very minimal in indica genotypes. Exploration of indica genetic diversity for grain yield traits will help us to generate improved knowledge on the physiological components of grain yield and thereby identify novel genes/alleles for rice improvement. In addition, rice genetic stocks differing in their grain number per panicle will allow us to understand the physiological relevance between sink size and dry matter partitioning. Hence, there is growing demand for discovering superior alleles of known candidate genes and also discovering novel genetic factors linked to grain yield traits in rice.

Mapping of genetic loci linked to any traits can be achieved through classical linkage mapping or through time saving Association mapping. Linkage disequilibrium based Association mapping strategies such as GWAS has several advantages over linkage mapping (Abdurakhmonov & Abdukarimov, 2008). Fundamental step in association mapping is to determine the genetic diversity and population structure of the study population (Kushwaha et al., 2017). GWAS has been successfully exploited in discovering genes specific markers associated with tiller number, yield and yield attributing traits and stress tolerance under different environmental conditions (Bhandari et al., 2020; Ren et al., 2021; Subedi et al., 2019).

Success and utility of GWAS depends on the availability of high density markers such as SNP markers. Rice being the model crop in monocots has been sequenced long back and recent efforts at IRRI, Philippines allowed generation of whole genome sequence information of 3024 diverse germplasm accessions from 89 countries (referred as 3 K Rice Panel) (RGP, 2014). Though such genomic dataset is readily available for public in Rice SNP-Seek Database (https://snp-seek.irri.org/), limited attempts have been made to exploit the genomic information for gene discovery (Abbai et al., 2019; Bhandari et al., 2020; Ren et al., 2021). Hence, it is assumed that exploitation of phenotypic diversity of 3 K panel and association mapping by utilizing the genomic information will identify novel genes/ alleles controlling grain yield traits in rice. In this study, efforts were made to analyze the genetic diversity and population structure of 217 diverse accessions of IRRI 3 K panel and a subset of 102 accessions was assembled for association mapping of panicle architecture and grain yield traits.

Materials and methods

Genetic materials and genomic data used

A set of 217 diverse rice genotypes of 3 K panel obtained from International Rice Research Institute, Philippines were used for genetic diversity and population structure analysis. CoreSNP Dataset (v0.4; 990 K) available at Rice SNP-Seek Database (http://snp-seek.irri.org) was used for analysis. Detailed information of this genotype file is described at (https://s3.amazonaws.com/3kricegenome/reduced/990k_3krg-snp-README.txt). Steps followed in performing genetic diversity, population structure and GWAS analysis are described in Fig. 1.

Fig. 1
figure 1

Step wise procedures followed in analyzing the genetic diversity, population structure and GWAS

Full size image

Population structure analysis and construction of a subset

Genotype data was filtered using TASSEL 5.0 (Bradbury et al., 2007) and used. SNPs with Minor Allele Frequency (MAF) ≥ 0.1 and heterozygous proportion ≤ 0.1 were retained, followed with pruning SNPs with minimum distance of 100,000 bp between the filtered SNPs. A set of 3449 SNPs covering all the 12 chromosomes were retained after filtering and pruning and used for inferring population structure. Population structure of 217 accessions was carried out using model-based maximum likelihood approach in ADMIXTURE 1.3 (Alexander et al., 2015) at different K values (K = 2 to 9). An optimum K with the least Cross-Validation (CV) error was fixed to retain pure accessions by eliminating admixtures.

Phenotyping of the subset for yield related traits

The subset of 102 rice accessions (Table 1) was evaluated for yield traits during October–January, 2020 at Paddy Breeding Station, Tamil Nadu Agricultural University, Coimbatore. Standard agronomic and nutrient management practices (150:50:50 N:P:K kg/ha) were followed. Observations on number of productive tillers/plant (NPT), number of primary branches/panicle (NPB), number of secondary branches/panicle (NSB), number of spikelets/panicle (NS) and single plant yield (SPY) were recorded. Observations were recorded in five randomly selected plants in a genotype. To assess the phenotypic variation and frequency distribution of evaluated traits in the subset, descriptive statistics and histogram were performed, respectively, using the Minitab 19.1(Allen, 2019).

Table 1 Details on the subset of 102 genetically pure rice accessions
Full size table

GWAS analysis

Genotypic data pertaining to the subset of 102 accessions was prepared as described above. Missing data were imputed using Beagle 4.1 (Browning & Browning, 2016). After filtering, a total of 101,388 polymorphic SNPs were selected for GWAS analysis of panicle and yield traits using GAPIT version 3 of Fixed And Random Model Circulating Probability Unification (FarmCPU) pipeline (Wang & Zhang, 2021). Manhattan Plots were constructed using qqman package (Turner, 2014) with a significant threshold of −log10 p value ≥ 4 to detect the significant marker trait associations. Genes closer to the identified SNPs (± 10 kb) were identified using The Rice Annotation Project-Database (RAP-DB) (Sakai et al., 2013) and Information Commons for Rice (IC4R) (Sang et al., 2020).

Results and discussion

Population structure of 217 rice accessions

Genetic diversity and population structure analysis is one of the basic requirements in any crop improvement programs. Population structure has been demonstrated to interfere with discovery of marker-trait-associations (Zhao et al., 2011). The present study was initiated with the assembly of 217 diverse rice accessions containing different sub-populations including 183 indica accessions (ind1A-19, ind1B-21, ind2-52, ind3-31, indx-60), 17 aus accessions, 11 japonica accessions (japx-4, temp-2, trop-5), 4 aromatic accessions and 2 admix accessions. All the 217 accessions were subjected to population structure analysis to eliminate the admixtures and to construct a subset for GWAS.

Model based maximum likelihood approach was used to infer population structure at different kinship (K) levels (K = 2 to 9) (Fig. 2). Though CV error gradually decreased with increasing K values, differences of CV error between each K were less and the least CV error was found at K = 8. Hence, K = 8 was considered as an optimum K to construct a subset of 102 accessions containing least admixtures providing a better statistical power for further GWAS analysis. At K = 2, the study population showed a clear distinctness between indica and other sub-populations. At K = 3, the study population showed distinctness within the indica sub-population where ind2 and other indica groups were distinctly separated into two sub-populations. On further increase of K, for example, at K = 5, the indica group of the study population showed distinct separation within their group. At K = 8, even within ind2 and ind3, there were distinct separation to different sub-populations. Lin et al. (2021) also inferred from population structure analysis of 2883 accessions that increased distinct separation of sub-populations can be observed when the K value increases. Though Lin et al. (2021) got an optimum K = 15 based on CV error, they fixed their own optimum at K = 5, due to prior knowledge of sub-populations in 3K-RGP and retained a total of 1378 accessions for GWAS. By following this approach, this study led to shortlisting of 102 accessions by retaining only pure types from 217 accessions with a minimum admixture of ≤ 0.2 ancestry proportions (0.0–1.0). Considering the need of pure types for perfect GWAS, this study has used K = 8 as optimum, admixtures were eliminated and pure types were retained (Table 1) and these 102 lines were used for GWAS.

Fig. 2
figure 2

Population structure of 217 accessions with K ranging from 2 to 9. Optimum K was shown at K = 8 based on CV error

Full size image

Phenotypic diversity for yield traits among the subset

The subset of 102 rice accessions constructed as above was evaluated for yield related traits. Descriptive statistics and frequency distribution of all traits were worked out to evaluate the trait diversity (Table 2 and Fig. 3). The panel showed diversification for all the evaluated traits. It was observed that single plant yield (SPY) exhibited maximum diversity in the study population followed by number of secondary branches per panicle (NSB), number of productive tillers per plant (NPT), number of spikelets per panicle (NS) and number of primary branches per panicle (NPB). Frequency distribution of all the five traits was found to be normal which indicated the suitability of the subset for GWAS.

Table 2 Descriptive statistics of yield related traits in the subset of 102 diverse rice lines
Full size table
Fig. 3
figure 3

Frequency distribution of yield related traits in the subset of 102 accessions

Full size image

GWAS identified novel genetic factors linked to yield related traits in rice

GWAS analysis was performed using the genotypic and phenotypic data of the subset of 102 genetically pure accessions so as to minimize false discovery rate (Korte & Farlow, 2013; Tian et al., 2008). GWAS using FarmCPU resulted in the identification of 42 SNPs significantly (threshold of –log10 p ≥ 4) linked to five different target traits. Among the 42 identified loci, 22 SNPs were linked to number of primary branches in the panicle (NPB) followed by 8 SNPs showing significant association with number of secondary branches per panicle (NSB), 6 SNPs with number of productive tillers per plant (NPT), 4 SNPs with number of spikelets per panicle (NS) and 2 SNPs with single plant yield (SPY) (Fig. 4 and Table 3). It was found that SPY and NS showed less SNPs association owing to their complex architecture as reported earlier (Korte & Farlow, 2013). Number of primary branches per panicle showed more number of associations due to the balanced distribution of NPB when compared to all other traits involved (Fig. 3) (Asimit & Zeggini, 2010). Amongst the 42 SNPs identified in this study, 20 SNPs were found to be located closer or co-localized with previously reported QTLs/genes related to panicle traits.

Fig. 4
figure 4

Manhattan plot showing the results of GWAS for yield traits in the 102 accessions

Full size image
Table 3 List of SNPs exhibiting significant marker-trait association for yield traits and their relatedness to previously reported loci
Full size table

SNP # 23943442 showed significant association with Number of secondary branches/ panicle. This SNP was found to be present within the candidate gene OsWD40-17 (LOC_Os01g42260) that encodes WD40/YVTN repeat-like-containing domain with varied functions including adaptor/regulatory modules in signal transduction, pre-mRNA processing and cytoskeleton assembly and acting as a transcriptional co-repressor modulating flower development (Gonzalez et al., 2007). Similarly SNP # 157578471 was identified as a significant SNP associated with number of productive tillers/plant. This encodes for OsMADS58 (LOC_Os05g11414), a MADS-box domain containing transcriptional factor and important for homeotic regulation in plants. This gene determines the floral meristem formation and involves in carpel development. SNP # 186260109 was found to be significantly associated with number of productive tillers/plant and found nearer to OsSTA173 (LOC_Os06g10140), which encodes Osfbl-29—F-Box Domain and LRR Containing Protein involved in the anther dehiscence, male fertility and pollen germination and determines the spikelet fertility (Ling et al., 2015).

Apart from this, the current study has identified 22 different novel genetic loci linked to various yield related traits. Phenotypic evaluation of the subset provided enough phenotypic variance for GWAS analysis. Among the 42 different genetic loci identified, 20 are co-localized with already reported QTLs/genes. These novel genetic factors may require validation through classical linkage mapping involving contrasting genotypes or validated in a large set of independent population. The subset of 102 rice accessions can be used for mapping agronomic and physiological traits such as nitrogen use efficiency, photosynthetic efficiency and so on. Superior genetic stocks and genes/alleles identified in this study will permit us to generate improved knowledge on the physiological and molecular basis of spikelet number per panicle in rice.