An IRAT109 NAM population: Genetic characterization and mapping utility for trait dissection in japonica rice (Oryza sativa L.)

Abstract

Complex agronomic traits in rice (Oryza sativa L.) are controlled by many genetic loci; however, dissection is often constrained in biparental mapping populations by limited recombination and narrow genetic diversity, and in diversity sets by spurious associations arising from cryptic relatedness. To overcome these limitations, we developed a nested association mapping (NAM) population of 422 recombinant inbred lines by crossing the upland tropical japonica IRAT109 with three genetically diverse japonica donors (Norin PL8, Langi, and RL11). Genotyping identified 4515 high‑quality polymorphic single nucleotide polymorphisms and revealed clear population stratification of the three subpopulations, with faster linkage disequilibrium decay (∼1.6 Mb) and broad allelic diversity, confirming that the genetic diversity of the NAM was sufficient for dissecting quantitative traits. Phenotyping under aerobic field conditions across two summer seasons for days to heading (DTH) and plant height (PH) recorded high heritabilities (0.96 for DTH; 0.92 for PH) and wide phenotypic ranges, confirming adequate phenotypic diversity for genomic analyses. Using multi‐locus genome‐wide association study (GWAS) models (Fixed and Random Model Circulating Probability Unification [FarmCPU] and Bayesian‐Information and Linkage‐Disequilibrium Iteratively Nested Keyway [BLINK]) alongside biparental linkage mapping, NAM identified nine additional loci undetected in biparental populations, improved mapping resolution by up to 6.3 Mb at specific loci, and revealed two quantitative trait loci (QTLs) co‐localizing with known genes controlling DTH and PH, while linkage mapping captured family‐specific allelic effects complementary to GWAS. Candidate genes were identified for the major‑effect QTLs qDTH6 (LOC_Os06g15370, LOC_Os06g16370, and LOC_Os06g16390) and qDTH7.1/qPH7.1 (LOC_Os07g15770), with sequence polymorphisms co‑segregating with donor alleles identified in GWAS, demonstrating the value of the IRAT109 NAM population for mapping and the discovery of loci underlying quantitative traits in rice.

Core Ideas

• A novel IRAT109 japonica rice nested association mapping (NAM) population (422 recombinant inbred lines [RILs]) with fast linkage disequilibrium (LD) decay (∼1.6 Mb) and broad allelic diversity.

• Broader genetic diversity in this NAM revealed nine quantitative trait loci (QTLs) missed by biparental populations and refined resolution by up to 6.3 Mb.

• Joint genome‐wide association study (GWAS) and linkage mapping in this NAM captured both shared and family‐specific alleles.

• Known candidate genes co‐locating within mapped QTL confirm robust detection in the population.

Plain Language Summary

Understanding the genes that control complex traits in rice is currently limited by traditional approaches. This study developed a novel population that maximizes genetic diversity to understand the genetic basis of flowering time and plant height. By combining DNA sequencing of the parents with genetic data from their offspring, it was demonstrated that the population contains sufficient genetic diversity to study quantitative traits in detail. Using genetic analysis methods, the study identified several regions of the genome affecting the studied traits, including both known genes and new candidate regions. This indicates that the design of the developed population was sufficient to pinpoint these regions much more precisely than traditional methods. Overall, this newly developed rice population is a valuable tool for discovering genes that influence key quantitative traits in rice.

Abbreviations

3K RGP

3K Rice Genomes Project

BLINK

Bayesian‐information and linkage‐disequilibrium iteratively nested keyway

BLUEs

best linear unbiased estimates

DTH

days to heading

FarmCPU

fixed and random model circulating probability unification

GWAS

genome‐wide association study

LD

linkage disequilibrium

MET

multi‐environment trial

NAM

nested association mapping

PC

principal component

PCA

principal component analysis

PH

plant height

QQ

quantile–quantile

QTL

quantitative trait locus

RIL

recombinant inbred line

SNP

single nucleotide polymorphism

Ti/Tv

transition/transversion

1. INTRODUCTION

Rice (Oryza sativa L.) is grown in 114 countries across 166.25 million ha (USDA FAS, n.d.) providing the caloric requirements of more than half of the world's population. As highlighted by Lenaerts et al. (2019), one of the most urgent challenges to food security is the accelerating pace of population growth. Addressing this demand necessitates continuous improvements in both potential yield and stress resilience, particularly in rice. However, many agronomic traits of economic significance in rice are quantitatively inherited and thus are controlled by multiple genes and shaped by complex interactions between genetic and environmental factors (Zaw et al., 2019). This complexity presents a significant obstacle to achieving consistent and sustainable genetic gains through conventional breeding approaches.

Traditional breeding approaches have relied primarily on the use of biparental populations for quantitative trait locus (QTL) mapping due to their simplicity and well‐defined genetic backgrounds (Zaw et al., 2019). While effective for detecting major QTL, biparental populations are constrained by limited recombination and narrow genetic diversity, often resulting in low mapping resolution and large linkage disequilibrium (LD) blocks (Gangurde et al., 2024; Xu et al., 2017). This limitation often necessitates additional rounds of fine mapping, extending the timeline and resource requirements for varietal improvement.

Alternatively, association mapping or genome‐wide association studies (GWASs) leverage diverse collections of unrelated genotypes to exploit historical LD and achieve higher mapping resolution (Ladejobi et al., 2016). Despite its advantages, GWAS can be confounded by population structure and cryptic relatedness, which may introduce spurious associations if not appropriately controlled (Sul et al., 2018).

To overcome the drawbacks of both biparental and association mapping populations, recent advances in plant genomics have prompted the adoption of multi‐parental mapping populations. Notably, nested association mapping (NAM) and multi‐parent advanced generation inter‐cross populations provide a powerful alternative by combining genetic diversity from multiple parental lines (Scott et al., 2020). NAM populations are established by crossing a diverse array of parents to one common reference parent, leading to a series of interconnected recombinant inbred lines (RILs) with minimized familial relatedness and enriched rare alleles (Gireesh et al., 2021). The design enhances allele sampling, breaks up LD, and generates novel recombination events, enabling both higher power and greater mapping resolution (Fragoso et al., 2017; Susmitha et al., 2023; Yu et al., 2008). This structure lays the foundation for high‐resolution linkage mapping while ameliorating the confounding effects in unrelated diverse panels.

NAM populations have demonstrated efficacy in dissecting the genetic architecture of complex traits across several crop species, including maize (Chen et al., 2019; Khatun et al., 2022; Ruiz‐Lozano et al., 2022), wheat (Alahmad et al., 2023; Bajgain et al., 2016; Chidzanga et al., 2022; Kidane et al., 2019; Wingen et al., 2017), sorghum (Bouchet et al., 2017; Olatoye et al., 2020), barley (Maurer et al., 2015; Saade et al., 2016; Schnaithmann et al., 2014), soybean (Beche et al., 2020; Maranna et al., 2019), and peanut (Gangurde et al., 2020). In rice, recent studies have underscored the promise of NAM for revealing novel loci associated with important agronomic traits such as days to heading (DTH) (Fragoso et al., 2017; Kitony et al., 2021), leaf blade length, stem length, stem width, panicle length, and panicle number (Nawade et al., 2024).

This study aims to (1) characterize the genetic diversity of the japonica lines used as NAM parents, (2) describe the genetic structure and phenotypic diversity of the IRAT109 NAM population, and (3) demonstrate the complementarity of the IRAT109 NAM population for mapping quantitative traits compared with traditional biparental populations. Beyond its genetic diversity, this population is uniquely poised for the discovery of loci governing low‐temperature tolerance, a trait for which japonica lines are particularly rich in beneficial alleles compared to indica backgrounds (Mackill & Lei, 1997). As such, the IRAT109 NAM resource represents not only a new platform for QTL dissection but also an invaluable tool in the pursuit of low‐temperature‐tolerant rice and traits of importance relevant to temperate environments.

2. MATERIALS AND METHODS

2.1. Plant materials

A structured japonica NAM, herein referred to as IRAT109 NAM, population of 422 genotypes was developed by crossing a common parent (IRAT109) with three different donor parents (Norin PL8, Langi, and RL11) and subsequently combining the resulting three RIL populations. The population consists of 174 RILs from the cross between IRAT109 and Norin PL8 (IRNO), 106 RILs from the cross between IRAT109 and RL11 (IRRL), and 142 RILs from the cross between IRAT109 and Langi (IRLA). These populations were developed through initial parental crosses followed by subsequent generations of selfing until the F₆ generation (Figure 1).

FIGURE 1.

Schematic representation of IRAT109 nested association mapping (NAM) population development showing the parental lines IRAT109, Langi, RL11, and Norin PL8 and their respective progenies IRLA (IRAT109 × Langi), IRRL (IRAT109 × RL11), and IRNO (IRAT109 × Norin PL8).

IRAT109 is an African deep‐rooting upland tropical japonica variety carrying qYMCT4 that decreases spikelet sterility when exposed to low‐temperature stress at the young microspore stage (Vinarao et al., 2022). It is also known to have strong shoots, high yield, and large seed size but with poor grain quality (Qu et al., 2008). Additionally, IRAT109 exhibits a narrow root cone angle and deeper rooting resulting in cooler canopy temperature and enhanced grain yield under aerobic conditions (Lou et al., 2015; Vinarao et al., 2021, 2023). Norin PL8 is a Japanese semi‐dwarf variety carrying Ctb‐1 and Ctb‐2 QTLs for low‐temperature tolerance at the young microspore stage (Saito et al., 2010, 2001, 1995). RL11 is a low‐temperature‐tolerant line identified from the cross between the Australian variety Reiziq, which has a moderate tolerance to low‐temperature stress at germination and seedling stage and is susceptible at booting and flowering stages, and the Chinese variety Lijiangheigu, which is tolerant across germination, seedling, booting, and flowering stages (Mitchell et al., 2016; Ye et al., 2009). Langi is a temperate Australian semi‐dwarf, soft‐cooking long‐grain variety with moderate tolerance to low temperature across germination, seedling, young microspore, and flowering (Ye et al., 2009). Other varieties and breeding lines with various tolerance and susceptibility to low temperature at different growth stages were also included as checks, such as Sherpa, an Australian variety with low‐temperature tolerance, and Matilda, a newly developed Australian variety with superior grain yield potential (Ward, 2021).

Core Ideas

• A novel IRAT109 japonica rice nested association mapping (NAM) population (422 recombinant inbred lines [RILs]) with fast linkage disequilibrium (LD) decay (∼1.6 Mb) and broad allelic diversity.

• Broader genetic diversity in this NAM revealed nine quantitative trait loci (QTLs) missed by biparental populations and refined resolution by up to 6.3 Mb.

• Joint genome‐wide association study (GWAS) and linkage mapping in this NAM captured both shared and family‐specific alleles.

• Known candidate genes co‐locating within mapped QTL confirm robust detection in the population.

2.2. Whole genome re‐sequencing of NAM parents

Whole genome resequencing for Langi and RL11 was carried out by submitting fresh leaf samples to the Australian Genome Research Facility, Australia, for sequencing. Library preparation was carried out using Illumina DNA Prep M Library Prep (Illumina), and the resulting libraries were sequenced using Illumina NovaSeq X (Illumina) paired‐end 150 bp reads. Whole genome resequencing of Norin PL8 was conducted as described by Vinarao et al. (2022). Briefly, high‐quality genomic DNA samples were extracted using a modified cetyltrimethylammonium bromide extraction method (Doyle & Doyle, 1987). The extracted DNA was sent to Macrogen Inc. for library preparation and sequencing. Library preparation was done using TruSeq Nano DNA 350 (Illumina), and the resulting libraries were sequenced using Illumina NovaSeq 6000 (Illumina) paired‐end 150 bp reads. Raw reads of IRAT109 were retrieved from the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA: ERS470547) publicly available through the 3K Rice Genomes Project (3K RGP) (Mansueto et al., 2017). All the sequences were pre‐processed for quality control, trimming, alignment, duplicate marking, and variant calling using Illumina's DRAGEN Bio‐IT platform v4.0.3 (v07.021.645.4.0.3) (Illumina) against the Oryza sativa MSU release 7 reference genome (Ouyang et al., 2006).

2.3. Single nucleotide polymorphism genotyping of the IRAT109 NAM population

Leaf tissue samples were collected, and the IRAT109 NAM lines were genotyped following the method described by Vinarao et al. (2021). This involved using the Diversity Arrays Technology genotyping‐by‐sequencing platform, followed by library sequencing with the Illumina Novaseq X. The sequences were then aligned to the Oryza sativa MSU release 7 reference genome (Ouyang et al., 2006) to identify single nucleotide polymorphisms (SNPs). SNP filtering was done to remove markers and genotypes with more than 25% missing data, more than 10% heterozygosity, and less than 5% minor allele frequency. Finally, random forest imputation was also performed to generate values for ∼6% missing calls using the “missForest” function in the missForest package (Stekhoven, 2025; Stekhoven & Bühlmann, 2012) in R (R Core Team, 2023).

2.4. Treatments and measurements

The IRAT109 NAM population, together with parents and checks, was grown at the University of Queensland, Gatton (27.5551° S, 152.3369° E) in the 2023/2024 and 2024/2025 summer cropping seasons, with a subset of lines not included in 2024/2025 due to failure of crop establishment in 2023/2024 or insufficient seed availability. An aerobic field was prepared with an oaten hay crop prior to the experiments and incorporated in the field 6–7 weeks before sowing. Dry seeds were drill‐seeded to a depth of 2–3 cm at a rate of 10 g/m², which aimed to establish plants at around 150 plants/m². Sowing was carried out on October 7, 2023 and October 16, 2024, corresponding to the 2024 and 2025 summer cropping seasons, respectively. Plots were 2 m in length with seven rows and an inter‐row spacing of 22 cm, resulting in a 3.08 m² plot size. The plots were arranged in a partially replicated design where the parents and check varieties were replicated three times, and the remaining entries were replicated either once or twice.

Basal fertilizer was applied 5 days prior to sowing at a rate of 400 kg/ha using Incitec Pivot's CK140S (23‐2‐18‐4) fertilizer. Prior to sowing, 15 kg/ha of zinc sulfate monohydrate was also applied and lightly incorporated. Pre‐emergent herbicide treatment was applied 5 days after initial irrigation and within 12 h prior to an irrigation event. The details of the treatment are as follows: (1) Stomp 440 (440 g/L Pendimethalin; CropCare), (2) Magister (480 g/L Clomazone; FMC Australia), (3) Gramoxone (250 g/L Paraquat, Syngenta). Sempra (750 g/kg Halosulfuron‐methyl; Nufarm) was used as post‐emergent herbicide treatment. Irrigation was carried out two to three times a week with 20–40 mm per irrigation event depending on the weather.

Days to 50% heading and plant height (PH) at maturity were collected. DTH was recorded as the number of days from sowing to when 50% of the plot had panicle emergence from the flag leaf. PH (cm) was measured from the base of the plant to the tip of the panicle (excluding awns).

2.5. Statistical analysis

2.5.1. Significance tests

Linear mixed models were fitted using the ASRreml (Butler et al., 2023) package in R (R Core Team, 2023) to estimate genetic effects and partition sources of variation for the traits under study. Following Gilmour et al. (1997) and Ovenden et al. (2017) with minor modifications, the model follows:

$$\mathbf{y}=\mathbf{X}\mathit{\tau }+{\mathbf{Z}}_{\mathbf{g}}\mathbf{g}+{\mathbf{Z}}_{\mathbf{u}}\mathbf{u}+\mathit{\eta }$$

where y is the (n × 1) vector of phenotypic observations across $p$ environments, each comprising $N$ plots. In this context, each environment was defined as a unique year‐by‐location combination, as the same physical location across different years represents distinct environments driven by year‐specific climatic and biotic conditions (Excellence in Breeding Platform, 2022). Accordingly, each year was treated as a single environment. τ is a (t × 1) vector of fixed effects, including linear trends across row and column with the associated design matrix X. g is a fixed component with associated design matrix Z _g used to model the genotype within environment effects, which combines the genotype and G × E interaction effects. u is a random component associated with design matrix Z _u that contains the experimental blocking structures used to capture extraneous variation such as field column, field row, replication, and individual plot‐level effects.

The residual error is η  = (η ₁,…, η _p), where for the jth experiment:

$${\mathit{\eta }}_j\sim N\left(0,{\mathrm{\sigma }}_j^2{\mathbf{R}}_{\mathbf{j}}\right)$$

where σ ² _j is the residual variance for the jth experiment and R_j is a spatial correlation matrix parameterized as a separable first‐order autoregressive process, AR1⊗AR1, applied across field rows and columns to account for spatial correlation among observations. For each analysis, model selection for nongenetic random effects was performed using log‐likelihood ratio tests relative to a full model containing all nongenetic terms, and effects were retained only when statistically significant. Fixed effects were assessed for significance using Wald F‐statistics.

2.5.2. Heritabilities and multi‐environment trial‐adjusted best linear unbiased estimates

Generalized heritability was calculated for each trait based on the formula proposed by Cullis et al. (2006) as follows:

$$H_2=1-\frac{{\mathrm{PEV}}_{{\mathrm{\Delta }}_{\mathrm{ij}}}^{\mathrm{BLUP}}}{2{\mathrm{\sigma }}_g^2}$$

where ${\mathrm{PEV}}_{{\mathrm{\Delta }}_{\mathrm{ij}}}^{\mathrm{BLUP}}$ is the prediction error variance matrix of the pairwise differences among BLUPS and 2σ ² _g is the variance attributed to differences between genotypes.

Multi‐environment trial (MET)‐adjusted best linear unbiased estimates (BLUEs) were extracted as the predicted genotype means from the final fitted single mixed model described above, with genotype fitted as fixed effects. This was done using the predict.asreml function in the ASRreml package. Prior to model fitting to generate BLUEs, residuals were assessed for normality to ensure the assumptions of the linear mixed model were met. The MET‐adjusted BLUEs were subsequently used in downstream GWASs and linkage mapping analyses.

2.5.3. Phylogenetic analysis

Phylogenetic analysis was carried out using common genetic markers identified between the whole‐genome resequencing data of the IRAT109 NAM parents (IRAT109, Norin PL8, Langi, and RL11) and the 3K RGP panel.

High‐density genotype data consisting of 4 million SNPs for the 3K RGP panel were obtained from the SNPseek database (https://brs-snpseek.duckdns.org/3kRG/). A representative subset was selected from the 3K RGP panel using the coreCollection package (Brouwer & Blok, 2022). This panel was selected such that the genetic diversity of the subset reflects the genetic diversity of the entire 3K RGP panel. Subsequently, a dissimilarity matrix was calculated using the poppr package (Kamvar et al., 2014), and a phylogenetic tree was constructed with the ape package (Paradis & Schliep, 2019). All the analyses were performed in R (R Core Team, 2023).

2.5.4. Genetic architecture and diversity analysis

Principal component analysis (PCA) and kinship relationships among the NAM lines were analyzed using the GAPIT package (Wang & Zhang, 2021). The first three principal components (PCs), selected based on the proportion of variance explained, were then included as covariates in the downstream GWAS models. PCA was visualized using the ggplot2 package (Wickham, 2016). The kinship matrix was visualized as a heatmap generated with the pheatmap package (Kolde, 2025). To illustrate the overlap of SNPs among NAM and RIL populations, intersection plots were constructed using UpSetR (Conway et al., 2017). The distribution of SNPs across the genome was examined using CMplot (LiLin‐Yin, 2024). Lastly, LD was calculated as the allele frequency correlation (r ²) for all pairwise SNP comparisons using Tassel 5.0 (Bradbury et al., 2007) in the unimputed genotype data and visualized using ggplot2 in R.

2.5.5. Linkage map development and mapping

Using the generated SNP data, polymorphic markers between IRAT109 and Norin PL8, IRAT109 and RL11, and IRAT109 and Langi were selected. These polymorphic SNPs were recoded in TASSEL 5.0 using an ABH genotype call, where each marker was classified as homozygous for the first parent (A), homozygous for the second parent (B), or heterozygous (H).

Marker genetic distance, in centiMorgans (cM), was calculated using the Kosambi map function of the ASMap package (Taylor & Butler, 2017) in R and the linkage map was estimated using the Lander–Green algorithm (Lander & Green, 1987). The SNP data were further processed to remove those with high missing rate, co‐location, evidence of genotyping errors, and outliers related to marker segregation distortion, proportion, double crossovers, and recombination numbers using the ASMap package.

Linkage mapping was done using the R/qtl package (Broman et al., 2003) in R. Conditional genotypic probabilities were calculated using the “calc.geno” function with step distance set to 1 cM. Genotypes were then simulated over the observed marker data using the “sim.geno” function with replications set to 32. Single and multiple QTL model analyses were done using “scanone” and “scantwo” functions, respectively, using the Haley–Knott regression method (Haley & Knott, 1992). The logarithm of odds threshold for QTL identifications was determined using 1000 permutations at 5% significance level. To identify multiple and interacting QTLs, “stepwiseqtl,” “refineqtl,” and “fitqtl” functions were used. Estimated effects and genetic variance explained (R ²) of each QTL were extracted. Lastly, Bayesian credible confidence intervals were determined using the “bayesint” function.

2.5.6. Genome‐wide association study

GWAS was conducted with MET‐adjusted BLUEs and imputed 4515 polymorphic markers using the “GAPIT” function in the GAPIT package (Wang & Zhang, 2021) in R using FarmCPU (Fixed and Random Model Circulating Probability Unification) (Liu et al., 2016) and BLINK (Bayesian‐Information and Linkage‐Disequilibrium Iteratively Nested Keyway) (Huang et al., 2018). For both models, PCs were set to 3 based on initial PCA results. These models were specifically designed to address confounding factors and enhance statistical power in association analyses as explained by Gong et al. (2025). The significance threshold for genome‐wide association was corrected for multiple testing using the Bonferroni correction method, with adjusted p values derived by dividing alpha (α = 0.05 and 0.0001) by the number of SNPs (N = 4515). The performance of different GWAS models was systematically evaluated using quantile–quantile (QQ) plots, which compare the expected versus observed −log₁₀(p) values for all tested markers. Lastly, QTL intervals were delineated based on local LD estimated from all pairwise SNP comparisons in the genomic region surrounding the peak SNP. Local LD analysis was performed in TASSEL version 5.0 using unimputed genotype data.

2.6. Candidate gene and sequence variant analysis

Whole genome sequence data from the NAM parents were used to investigate sequence variation within genes associated with DTH and PH across the identified major‐effect loci (R ² > 10%). Genes residing within these loci were annotated using the Rice Annotation Project Database (RAP‐DB; https://rapdb.dna.naro.go.jp/) (Sakai et al., 2013). Sequence variation among the NAM parents was assessed using TASSEL version 5.0. Only variants that co‐segregated with the donor alleles identified in the GWAS analysis were considered valid for further interpretation.

3. RESULTS

3.1. Sequencing metrics and phylogenetic analysis of NAM parents

High‐quality whole genome sequencing data were obtained from four parents, with mapping statistics summarized in Table 1 for the parents (Langi, IRAT109, Norin PL8, and RL11). The total mapped reads vary among parents, with RL11 having the highest (333 M) and Norin PL8 having the lowest (77 M).

TABLE 1.

Key sequencing and variant calling metrics of IRAT109 nested association mapping (NAM) parents.

Genotype	Total reads mapped	Aligned reads (%)	Properly mapped reads (%)	Total variants	Transition/transversion ratio
Langi	150,022,608	98.44	97.07	1,256,812	2.39
IRAT109	107,543,727	98.32	87.34	1,266,581	2.40
Norin PL8	76,744,003	99.14	97.79	364,224	2.19
RL11	333,137,426	97.04	95.77	807,064	2.30

Meanwhile, all parents exhibited high alignment rates (∼97%–99%) with Norin PL8 (99.14%) having the highest alignment proportions. The properly mapped read percentages varied more substantially, from 87.34% in IRAT109 to 97.79% in Norin PL8. IRAT109 showed a notably lower percentage of properly mapped reads compared to the other parents, which all exceeded 95%. Variant counts also varied widely, with IRAT109 (1.27 M) exhibiting the highest variant counts together with Langi (1.26 M), while Norin PL8 (364K) and RL11 (807K) have the lowest variant counts. The transition/transversion (Ti/Tv) ratios for SNP calls were within the expected range (2.19–2.4) for high‐quality rice genome data.

PCA was performed to compare the genetic diversity of the selected subset (red dots) with the full 3K RGP panel (gray dots) (Figure 2A). The even distribution of selected lines demonstrated that the selection strategy retained the underlying population structure and genetic diversity of the 3K RGP panel. A neighbor‐joining phylogenetic tree constructed from the common 347,092 SNPs between NAM parents and selected the 3K RGP panel (Figure 2B) grouped the NAM parental lines across different subpopulations within the japonica group, including both temperate and tropical japonica.

FIGURE 2.

(A) Principal component analysis plot of the selected subset compared to the 3K RGP (3K Rice Genomes Project) panel showing the retention of genetic diversity in the selected subset. (B) Neighbor‐joining phylogenetic tree of IRAT109 nested association mapping (NAM) parents with a representative subset of the 3K RGP based on 347,092 single nucleotide polymorphisms (SNPs) across all 12 chromosomes.

3.2. Genetic structure and diversity of NAM population

A total of 4515 high‐quality polymorphic SNPs were identified and mapped across the 12 rice chromosomes in 1 Mb windows across the NAM population (Figure 3A). The SNPs exhibited a nonuniform distribution, with certain chromosomal regions with higher marker densities. PCA of the IRAT109 NAM population, using the polymorphic SNP markers, revealed pronounced population structure, with clear separation of three subpopulations (IRNO, IRLA, and IRRL) and distinct clustering of RILs by subpopulation (Figure 3B). The first two PCs (PC1 = 14.36% and PC2 = 8.36%) account for 22.72% of the total genetic variation and distinctly separated the three subpopulations according to their subpopulation of origin. Moreover, the NAM parent lines (Langi, IRAT109, RL11, and Norin PL8) were clearly differentiated, which highlighted the genetic diversity introduced by the NAM design.

FIGURE 3.

(A) Distribution of 4515 polymorphic single nucleotide polymorphisms (SNPs) within 1 Mb window size across the 12 rice chromosomes. (B) Principal component (PC) analysis showing the population structure in the IRAT109 nested association mapping (NAM) population based on PC1 and PC2 using 4515 SNP markers. Each point represents an individual recombinant inbred line (RIL), colored by subpopulation (IRNO, IRLA, and IRRL). NAM parent lines (Langi, IRAT109, RL11, and Norin PL8) are individually labeled. (C) Scree plot of the variance explained by each of the first 10 principal components. The optimal number of PC (k = 3) is indicated by a red circle. (D) Heatmap of kinship matrix showing the degree of relatedness of IRAT109 NAM lines. (E) Intersection plot of SNP distribution between NAM, IRLA, IRNO, and IRRL populations. (F) Linkage disequilibrium decay plot of NAM, IRLA, IRNO, and IRRL populations.

The scree plot (Figure 3C) indicated that the first three PCs accounted for the majority of genetic variance within the population, with an optimal number of PCs (k = 3) corresponding to the three subpopulations. Furthermore, kinship analysis (Figure 3D) revealed distinct patterns of relatedness among the IRAT109 NAM lines. Higher degrees of relatedness were observed within subpopulations, while lower relatedness was evident between subpopulations that corroborate the PCA results.

Intersection analysis of SNP distribution among the RIL populations and the overall NAM population (Figure 3E) identified both shared and unique SNPs, in which the NAM population utilized 2214 additional unique polymorphic SNPs over the biparental populations. This represents a 3.9‐ to 5.6‐fold increase in markers when using NAM compared to individual biparental populations. LD decay (Figure 3F) in the NAM population also occurred more rapidly than in the individual RIL populations, with r ² declining to half of its maximum at approximately 1.6 Mb. In comparison, the decay to half was about 2.2 Mb in IRNO, 2.0 Mb in IRLA, and 1.8 Mb in IRRL.

3.3. Phenotypic characterization of NAM population

Significant variation was observed in DTH and PH across the NAM population (Table 2). DTH ranged from 66 to 136 days with a population mean of 90 days, while PH ranged from 48 to 143 cm with a mean of 98 cm. The phenotypic ranges for DTH (70‐day spread) and PH (95‐cm spread) exceeded the values recorded from the parental lines, which spanned 69–101 days for DTH and 73–108 cm for PH, indicating transgressive segregation in the population. Phenotypic ranges also differed among subpopulations for both traits. IRNO exhibited the widest range for both DTH (66–136 days) and PH (48–143 cm), and IRRL showed intermediate variation (DTH: 68–131 days; PH: 60–135 cm), while IRLA displayed the narrowest ranges (DTH: 79–109 days; PH: 71–120 cm).

TABLE 2.

Multi‐environment trial (MET) summary statistics (±standard error), heritability estimates, and year‐to‐year correlation for days to heading (DTH) and plant height (PH, cm) in IRAT109 nested association mappings (NAMs), parents, and checks evaluated under aerobic field conditions at the University of Queensland (UQ), Gatton, in the 2024 and 2025 summer cropping seasons.

	DTH	PH
NAM
Min	66	48
Max	136	143
Mean	90***	98***
Parents and checks
IRAT109	98 ± 1.35	108 ± 2.42
Langi	101 ± 1.35	96 ± 2.56
Norin PL8	69 ± 1.35	73 ± 2.40
RL11	79 ± 1.35	84 ± 2.53
Sherpa	99 ± 1.48	87 ± 2.37
Reiziq	89 ± 1.35	81 ± 2.40
Matilda	92 ± 1.35	87 ± 2.56
Heritability	0.96	0.92
Year‐to‐year correlation	0.88	0.81

***p < 0.001.

The performance of the NAM population was benchmarked against four parents and three check varieties. Among parents, Langi was the slowest to reach heading (101 days), followed by IRAT109 (98 days), which was also the tallest genotype (108 cm). In contrast, Norin PL8 was the earliest to reach heading (69 days) and was the shortest genotype among the parents (73 cm). RL11 displayed intermediate heading (79 days) and height (84 cm). Among the check varieties, Reiziq (89 days) and Matilda (92 days) headed closely with the NAM population, while Sherpa headed later (99 days). All three check varieties, Reiziq (81 cm), Matilda (87 cm), and Sherpa (87 cm), recorded shorter PH than the NAM population mean. Broad‐sense heritability was high for both traits, with estimates of 0.96 for DTH and 0.92 for PH. Strong year‐to‐year correlations were also observed between 2024 and 2025 for DTH (r = 0.88) and PH (r = 0.81).

3.4. Genomic regions controlling DTH and PH identified using NAM GWAS

The performance of different GWAS models, FarmCPU and BLINK, was systematically evaluated using QQ plots (Figure 4). For both models, the majority of markers in the GWAS analysis showed observed −log₁₀(p) values that closely aligned with the expected diagonal. However, there were notable upward deviations at higher observed −log₁₀(p) values, indicating true association signals. Additionally, both models demonstrated comparable performance in detecting genetic associations.

FIGURE 4.

Quantile–quantile (QQ) plots comparing expected and observed −log₁₀(p) values across genome‐wide association study (GWAS) models (Fixed and Random Model Circulating Probability Unification [FarmCPU] and Bayesian‐Information and Linkage‐Disequilibrium Iteratively Nested Keyway [BLINK]) for (A) days to heading and (B) plant height.

Using both GWAS models, nine QTLs associated with DTH were identified across chromosomes 1, 3, 5, 6, 7, and 12 (Figure 5; Table 3), revealing a predominantly oligogenic trait architecture in which a small number of major‐effect loci account for the bulk of explained variation. Three QTLs, qDTH3.1, qDTH6, and qDTH7.1, acted as major‐effect loci, collectively explaining 64.47% of the total phenotypic variance captured, while the remaining six loci each explained less than 5.50% and are considered minor‐effect loci. The QTL qDTH7.1 showed the largest negative effect on heading (−17.2 days) and the highest phenotypic variation (33.88). In contrast, qDTH3.1 had the largest effect to delaying heading (+12.5 days). This was followed by qDTH6, which increased heading time by approximately 10.4 days. The remaining QTL exhibited smaller negative allelic effects on DTH, ranging from −3.6 to −5.2 days, each contributing modestly to the overall variation. Notably, the major‐effect loci qDTH3.1, qDTH6, and qDTH7.1 and minor‐effect loci qDTH1.3, qDTH5.1, and qDTH7.2 were detected using both GWAS models, whereas qDTH1.1 and qDTH12.1 were detected only by FarmCPU and qDTH1.2 only by BLINK. Finally, QTL interval sizes varied from 0.30 to 1.85 Mb, with qDTH1.1 and qDTH1.2 defined by the narrowest intervals and qDTH6 and qDTH1.3 by the widest.

FIGURE 5.

Manhattan plots using Fixed and Random Model Circulating Probability Unification (FarmCPU) and Bayesian‐Information and Linkage‐Disequilibrium Iteratively Nested Keyway (BLINK) for (A) days to heading and (B) plant height. The blue solid and broken horizontal lines represent the thresholds of 7.65 and 4.96, which are the significance thresholds based on Bonferroni correction using α values 0.0001 and 0.05, respectively. Red vertical lines represent a significant single nucleotide polymorphism (SNP) detected by both FarmCPU and BLINK models. Major loci (R ² > 10%) are labeled.

TABLE 3.

Genomic regions significantly associated with days to heading (DTH) and plant height (PH) evaluated under aerobic field conditions detected using at least one genome‐wide association study (GWAS) model in nested association mapping (NAM). For regions identified by both GWAS models, the reported −log₁₀(p), R ², and effects were taken from the GWAS model with the higher R ².

Trait	QTL	Model	Chr	Pos (Mb)	Interval (Mb)	−log10(p)	R 2	Effect	Donor allele
DTH	qDTH1.1	FarmCPU	1	6.71	6.56–6.86	5.19	2.45	−3.7	Norin PL8, RL11, Langi
	qDTH1.2	BLINK	1	7.42	7.24–7.60	5.01	2.07	−3.6	Norin PL8, RL11, Langi
	qDTH1.3	BLINK, FarmCPU	1	31.78	30.85–32.70	6.32	3.18	−4.9	Norin PL8, RL11
	qDTH3.1	BLINK, FarmCPU	3	1.03	0.73–1.33	35.37	19.20	12.5	Norin PL8, RL11, Langi
	qDTH5.1	BLINK, FarmCPU	5	27.26	26.38–28.14	6.81	3.43	−4.2	IRAT109
	qDTH6	BLINK, FarmCPU	6	9.54	8.60–10.48	12.58	11.39	10.4	Norin PL8, RL11
	qDTH7.1	BLINK, FarmCPU	7	9.23	8.98–9.48	37.84	33.88	−17.2	Norin PL8
	qDTH7.2	BLINK, FarmCPU	7	28.97	28.70–29.24	6.11	5.29	−5.2	Norin PL8, RL11
	qDTH12.1	FarmCPU	12	16.71	15.90–17.52	6.15	2.17	−4.0	Norin PL8, RL11
PH	qPH1.1	BLINK	1	6.87	6.72–7.02	6.86	1.31	−5.0	Norin PL8, RL11, Langi
	qPH1.2	BLINK, FarmCPU	1	37.14	36.59–37.69	24.11	20.86	16.3	Norin PL8
	qPH1.3	BLINK	1	41.97	41.47–42.47	7.05	1.34	5.2	Norin PL8, RL11, Langi
	qPH3.1	BLINK, FarmCPU	3	1.03	0.73–1.33	32.87	14.76	14.3	Norin PL8, RL11, Langi
	qPH6.1	BLINK, FarmCPU	6	31.00	30.17–31.83	7.00	4.15	−6.7	Norin PL8, Langi
	qPH7.1	BLINK, FarmCPU	7	9.23	8.98–9.48	8.77	18.98	−11.6	Norin PL8
	qPH7.3	BLINK	7	15.44	15.28–15.60	5.52	2.55	−6.8	Norin PL8, RL11
	qPH7.2	BLINK, FarmCPU	7	28.97	28.70–29.24	21.56	17.81	−12.9	Norin PL8, RL11
	qPH9.1	FarmCPU	9	17.19	16.22–18.17	5.45	1.86	−4.4	IRAT109
	qPH12.1	BLINK	12	25.14	24.95–25.43	5.32	2.31	−4.2	Norin PL8, RL11, Langi

Note: Chr, chromosome; Pos, position in Mb; −log₁₀(p), Log‐scaled p‐value; R ², phenotypic variation; Effect, estimated allelic effects of the donor allele.

Abbreviations: BLINK, Bayesian‐Information and Linkage‐Disequilibrium Iteratively Nested Keyway; FarmCPU, Fixed and Random Model Circulating Probability Unification; QTL, quantitative trait locus.

For PH, 10 QTLs were identified across chromosomes 1, 3, 6, 7, 9, and 12, following a similarly oligogenic architecture dominated by four major‐effect loci (Figure 5; Table 3). The loci qPH1.2, qPH3.1, qPH7.1, and qPH7.2 were classified as major‐effect loci, each explaining more than 10% of phenotypic variation and collectively accounting for 72.41% of the phenotypic variance captured across all PH loci. qPH1.2 accounted for the largest proportion of variation (20.86%) and increased PH by approximately 16.3 cm. Meanwhile, qPH7.1, which explained 18.98% of the variation, and qPH7.2, which accounted for 17.81%, reduced PH by 11.6 and 12.9 cm, respectively. Notably, qPH3.1 recorded the highest −log₁₀(p) value, despite a smaller effect size of 14.76% of the phenotypic variation and increasing PH by 14.3 cm. All four major‐effect loci were detected by both FarmCPU and BLINK. Among the minor‐effect QTL, qPH6.1 was also detected using both models, while qPH1.1, qPH1.3, qPH7.3, and qPH12.1 were detected only by BLINK, and qPH9.1 only by FarmCPU. As with DTH, QTL interval sizes varied from 0.30 to 1.95 Mb, with qPH1.1 and qPH7.3 showing the narrowest intervals, and qPH9.1 and qPH6.1 the widest.

Notably, four genomic regions were associated with both DTH and PH, suggesting either pleiotropy, shared genetic control, or tight linkage between loci. For qDTH1.1/qPH1.1, alleles from Norin PL8, RL11, and Langi were associated with earlier heading and reduced PH. Similarly, the Norin PL8 allele at qDTH7.1/qPH7.1 and the Norin PL8 and RL11 alleles at qDTH7.2/qPH7.2 conferred earlier heading and shorter height. In contrast, at qDTH3.1/qPH3.1, alleles from Norin PL8, RL11, and Langi were associated with delayed heading and increased PH. Across all four regions, the same directionality of effects of DTH and PH indicates a possible shared genetic control that confers early heading with short PH and delayed heading with taller PH.

3.5. Genomic regions identified for DTH and PH using biparental linkage mapping

Biparental linkage mapping revealed differences in the genetic architecture of DTH and PH across the three populations. For DTH, the number of QTL detected varied considerably, with IRRL exhibiting the most complex architecture (nine loci), compared with five in IRNO and three in IRLA (Tables 4 and 5), suggesting that the genetic basis of flowering time differs substantially among crosses. For PH, numbers were more comparable across populations (five in IRLA, six in IRNO, and five in IRRL). Most of the loci were population‐specific; however, three DTH loci (qDTH3.1, qDTH4.2, and qDTH6) and two loci (qPH3.1 and qPH7.2) were consistently detected in at least two populations, while the other regions were unique in each biparental population, indicating partially shared genetic control across genetic backgrounds. The phenotypic variation explained by these QTLs ranged from 2.92% to 46.40% for DTH and between 5.18% and 29.90% for PH, with donor alleles conferring both increases and decreases in trait values, reflecting bidirectional selection potential within each of the biparental populations. Notably, for qDTH3.1 and qDTH6, alleles from both Norin PL8 and RL11 contributed to delayed heading; for qDTH4.2, alleles from Langi and RL11 contributed to earlier heading; and for qPH3.1 and qPH7.2, alleles from both Norin PL8 and RL11 contributed to increased and reduced PH, respectively.

TABLE 4.

Genomic regions significantly associated with days to heading (DTH) evaluated under aerobic field conditions detected using biparental linkage mapping.

Pop	QTL	Chr	Pos (Mb)	Interval (Mb)	LOD	R 2	Effect
IRLA	qDTH3.2	3	12.61	11.07–14.93	9.01	17.17	5.4
	qDTH4.2	4	31.19	30.04–31.19	18.22	41.29	−8.8
	qDTH12.2	12	26.77	25.94–26.90	8.01	14.99	5.2
IRNO	qDTH3.1	3	1.39	1.05–1.50	17.78	20.09	13.5
	qDTH5.2	5	22.24	20.83–27.07	5.78	5.53	−7.6
	qDTH6	6	9.34	9.26–11.06	10.24	10.41	3.5
	qDTH7.1	7	9.23	7.78–9.23	32.89	46.40	−20.2
	qDTH7.2	7	27.69	22.09–28.97	6.98	6.78	−7.8
IRRL	qDTH2.1	2	2.94	2.94–2.98	12.12	9.33	18.5
	qDTH2.2	2	3.85	3.27–6.58	8.68	6.09	−15.9
	qDTH3.1	3	1.05	0.93–1.05	22.14	22.61	12.4
	qDTH3.3	3	31.49	31.49–32.97	16.86	14.79	−9.4
	qDTH4.1	4	5.21	0.99–5.81	5.72	3.71	−5.1
	qDTH4.2	4	30.32	30.04–30.87	11.81	9.02	−8.9
	qDTH6	6	10.19	8.96–11.84	26.70	31.20	20.2
	qDTH11	11	22.24	20.91–24.34	4.63	2.92	−4.5
	qDTH12.1	12	15.46	14.86–16.99	6.32	4.17	−6.8

Note: Pop, population; Chr, chromosome; Pos, position in Mb; R ², phenotypic variation; Effect, estimated allelic effects of the donor (Langi for IRLA, Norin PL8 for IRNO, and RL11 for IRRL) with respect to IRAT109.

Abbreviations: LOD, logarithm of odds; QTL, quantitative trait locus.

TABLE 5.

Genomic regions significantly associated with plant height (PH) evaluated under aerobic field conditions detected using biparental linkage mapping.

Pop	QTL	Chr	Pos (Mb)	Interval (Mb)	LOD	R 2	Effect
IRLA	qPH1.4	1	4.57	4.18–6.56	3.75	5.97	−4.5
	qPH4	4	30.04	22.64–33.60	4.54	7.35	−5.5
	qPH5	5	29.41	28.80–29.41	8.20	14.17	2.1
	qPH6.3	6	25.39	25.03–26.85	4.63	7.50	−5.3
	qPH11	11	4.55	4.41–5.89	13.94	26.79	−4.6
IRNO	qPH1.2	1	37.14	37.14–39.37	20.39	22.23	19.3
	qPH3.1	3	1.39	0.79–1.50	11.22	10.75	12.1
	qPH6.4	6	30.03	25.53–31.19	6.40	5.74	−10.5
	qPH7.1	7	9.23	7.78–11.54	21.26	23.47	−18.7
	qPH7.2	7	28.92	27.84–28.97	12.06	11.68	−12.5
	qPH12.2	12	24.05	23.07–24.75	3.98	3.45	−6.9
IRRL	qPH3.1	3	1.05	0.93–2.11	14.98	29.90	18.7
	qPH3.2	3	34.14	31.49–34.69	3.51	5.18	−9.8
	qPH6.2	6	13.33	10.65–19.91	7.05	11.39	13.8
	qPH7.2	7	28.39	27.70–28.97	13.33	25.43	−16.9
	qPH9.1	9	15.08	14.11–17.74	4.80	7.31	13.5

Abbreviations: Pop, population; Chr, chromosome; Pos, position in Mb; R ², phenotypic variation; Effect, estimated allelic effects of the donor (Langi for IRLA, Norin PL8 for IRNO, and RL11 for IRRL) with respect to IRAT109.

Abbreviations: LOD, logarithm of odds; QTL, quantitative trait locus.

Across the three biparental populations, several genomic regions influenced both DTH and PH, indicating either pleiotropy, shared genetic control, or tight linkage between loci governing flowering time and plant stature. Within each colocalized region, the direction of the donor allele effect was proportional across traits where alleles associated with delayed heading were consistently associated with increased PH and those promoting earlier heading with reduced height. This pattern was observed across all three populations, including qDTH4.2/qPH4 in IRLA, qDTH3.1/qPH3.1, qDTH7.1/qPH7.1, and qDTH7.2/qPH7.2 in IRNO, and qDTH3.1/qPH3.1, qDTH3.3/qPH3.2, and qDTH6/qPH6.2 in IRRL. The consistency of this relationship across the biparental populations suggests a functional or regulatory coupling between the two traits in these populations.

3.6. Comparison of genomic regions identified using NAM GWAS and biparental linkage mapping

Comparison between using traditional linkage mapping in biparental populations (IRLA, IRNO, and IRRL) and GWAS in NAM revealed substantial overlaps in the detected genomic regions, as well as differences that underscore the utility of the NAM approach.

For DTH, NAM detected nine loci spanning chromosomes 1, 3, 5, 6, 7, and 12 (Table 3). IRLA and IRNO each identified considerably fewer loci, three and five, respectively (Table 4). While NAM and IRRL each detected nine QTLs for DTH, the NAM population revealed more major shared regions with other populations such as those on chromosomes 3, 6, and 7. For PH, the NAM population identified the most number of QTLs, detecting 10 loci across chromosomes 1, 3, 5, 6, 7, 9, and 12 (Table 3). In contrast, the biparental populations detected fewer loci, five in IRLA, six in IRNO, and five in IRRL (Table 5).

Across both traits, NAM also uncovered several loci that were not detected in all biparental populations, such as those on chromosome 1 (qDTH1.1, qDTH1.2, qDTH1.3, qPH1.1, and qPH1.3), chromosome 5 (qDTH5.1), chromosome 6 (qPH6.1), chromosome 7 (qPH7.3), and chromosome 12 (qPH12.1) (Figure 6). Notably, allelic effects in these regions were contributed by several different donors, reflecting the enhanced diversity of the NAM panel and its ability to detect genomic regions absent from the biparental populations.

FIGURE 6.

Overview of quantitative trait locus (QTL) overlap for days to heading and plant height across nested association mapping (NAM), IRLA, IRNO, and IRRL populations. Colored boxes to the left and right of each chromosome indicate days to heading (DTH) and plant height (PH) QTL, respectively, with lengths proportional to the QTL interval size.

In addition to identifying more regions overall, the NAM population also recovered several QTLs that overlapped with those detected in the subpopulations. Figure 6 showed the DTH QTL on chromosomes 3, 6, 7, and 12 that were detected in both the NAM population and at least one of the biparental populations. For PH, shared regions were observed on chromosomes 1, 3, 7, and 9. However, it is worth noting that several loci were also exclusively detected in biparental populations. For DTH, these include a locus on chromosome 3 identified using IRLA, a locus on chromosome 5 using IRNO, and multiple loci on chromosomes 2, 3, 4, and 11 using IRRL. Similarly, for PH, exclusively detected loci span chromosomes 1, 4, 5, 6, and 11 in IRLA, chromosomes 6 and 12 in IRNO, and chromosomes 4 and 6 in IRRL.

Investigation of the common genomic regions revealed that QTL intervals were generally narrower in the NAM population than in the biparental populations. qDTH7.2 on the long arm of chromosome 7 was reduced from a 6.88 Mb interval in IRNO to 0.54 Mb in NAM. Similarly, qDTH7.1 on the short arm of chromosome 7 was refined from 1.45 Mb in IRNO to 0.50 Mb in NAM. On chromosome 12, qDTH12.1 was narrowed from 2.13 Mb in IRRL to 1.62 Mb in NAM. Two regions, however, showed slightly different patterns, indicating that mapping resolution varied depending on the genomic region. On chromosome 6, qDTH6 spanned 1.88 Mb in NAM, which was comparable to its interval in IRNO (1.80 Mb) but narrower than in IRRL (2.88 Mb). Conversely, qDTH3.1 on chromosome 3 was detected as a 0.60 Mb interval in NAM, but IRNO and IRRL resolved the same locus to just 0.45 and 0.12 Mb.

A comparable trend was observed for PH, with narrower intervals consistently identified in NAM. On chromosome 1, qPH1.2 was refined from 2.23 Mb in IRNO to 1.10 Mb in NAM. On chromosome 3, qPH3.1 was reduced to 0.60 Mb in NAM, compared with intervals of 1.18 Mb in IRRL and 0.71 Mb in IRNO. On the short arm of chromosome 7, qPH7.1 was refined from 3.76 Mb in IRRL to 0.50 Mb in NAM, while qPH7.2 on the long arm was reduced to 0.54 Mb in NAM compared with 1.13 Mb in IRNO and 1.27 Mb in IRRL. Finally on chromosome 9, qPH9.1 was narrowed from 3.63 Mb in IRRL to 1.95 Mb in NAM.

Two regions merit additional consideration. First, overlapping DTH loci on the long arm of chromosome 5 were distinguished as separate QTL due to opposing allelic effects of IRAT109. In qDTH5.1 (NAM), the IRAT109 allele was associated with earlier heading, whereas in qDTH5.2 (IRNO), it was associated with delayed heading (Tables 3 and 4). Second, although qPH6.4 in IRNO (25.53–31.19 Mb) overlaps with both qPH6.1 (30.17–31.83 Mb) in NAM and qPH6.3 (25.03–26.85 Mb) in IRLA, the lack of overlap between qPH6.1 and qPH6.3 indicates that these loci likely represent distinct QTL.

3.7. Candidate genes identified using parental sequence variant analysis

Candidate gene analysis within the major‐effect QTL intervals identified several plausible genes for heading date and PH in the population. Using the RAP‐DB annotation, three candidate genes related to heading date were identified within the qDTH3.1 interval, six within qDTH6, and three within qDTH7.1. For PH, four candidate genes were identified within qPH1.2, five within qPH3.1, two within qPH7.1, and five within qPH7.2. However, only candidates within qDTH6, qDTH7.1, and qPH7.1 have variants that co‐segregated with the donor alleles identified in the GWAS analysis (Table 6). It must be noted, however, this candidate gene set represents a prioritized shortlist for downstream functional characterization, rather than definitive evidence of causality.

TABLE 6.

Sequence variants within candidate genes underlying major‐effect quantitative trait locus (QTL) (R ² > 10%).

QTL	Candidate gene	Donor	Alternative	Variant position	Variant	Type
qDTH6	LOC_Os06g15370	Norin PL8, RL11	IRAT109, Langi	8742930	A/G	SNP
	LOC_Os06g16370	Norin PL8, RL11	IRAT109, Langi	9336782	AA/GC	MNP
				9336938	G/A	SNP
				9336964	A/C	SNP
				9336967	A/C	SNP
				9336985	A/G	SNP
				9337148	A/G	SNP
				9338004	–/T	INDEL
				9338068	C/A	SNP
				9338330	A/G	SNP
	LOC_Os06g16390	Norin PL8, RL11	IRAT109, Langi	9354533	T/G	SNP
				9355564	T/C	SNP
				9357118	C/T	SNP
				9357155	GC/CT	MNP
				9357300	T/C	SNP
				9357456	T/A	SNP
				9357516	C/T	SNP
				9357730	G/A	SNP
				9361740	A/G	SNP
				9361807	–/G	INDEL
				9362026	G/A	SNP
				9362086	–/A	INDEL
qDTH7.1/qPH7.1	LOC_Os07g15770	Norin PL8	IRAT109, Langi, RL11	9154664	C/A	SNP

Note: Donor represents parents identified as the effect source through genome‐wide association study (GWAS); Alternative represents the remaining parents. Variant notation follows Donor/Alternative allele format.

Abbreviations: INDEL, insertion‐deletion; MNP, multiple nucleotide polymorphism; SNP, single nucleotide polymorphism.

Within qDTH6, three genes contained polymorphisms that differentiated the expected donor and alternative parent genotypes: LOC_Os06g15370, LOC_Os06g16370, and LOC_Os06g16390. These genes fall within the chromosome 6 interval associated with delayed heading, where the donor alleles from Norin PL8 and RL11 increased DTH by about 10.4 days. LOC_Os06g15370 corresponds to NITRATE TRANSPORTER 1/PEPTIDE TRANSPORTER 3.1 (OsNPF3.1), which contained a single SNP at 8,742,930 bp, with Norin PL8 and RL11 carrying the donor allele relative to IRAT109 and Langi. A higher density of polymorphisms was observed in LOC_Os06g16370, which corresponds to HEADING DATE 1 (Hd1), which carried one multiple nucleotide polymorphism (MNP), several single SNPs, and one insertion‐deletion (INDEL). LOC_Os06g16390 that corresponds to SET DOMAIN GROUP 711 (SDG711) also showed substantial allelic variation, including SNPs, MNP, and INDELs.

In contrast to several candidate genes and variations identified within qDTH6, both qDTH7.1 and qPH7.1 showed the expected variant co‐segregation in a single candidate gene, LOC_Os07g15770 (grain number, PH, and heading date 7; Ghd7). Within this gene, only one SNP was identified at 9,154,664 bp. This region corresponds to the major‐effect locus on chromosome 7 that reduced heading time by 17.2 days and PH by 11.6 cm, with the favorable donor allele contributed by Norin PL8.

4. DISCUSSION

This study demonstrates the development of a novel IRAT109 NAM population derived from a diverse set of japonica rice parents, representing a significant methodological leap forward in rice genetic resources. Unlike traditional biparental populations, the IRAT109 NAM design deliberately incorporated multiple, genetically diverse parental lines spanning both temperate and tropical japonica subpopulations, thereby capturing a broad spectrum of allelic variation across the genome. This deliberate structural diversity, confirmed through phylogenetic analysis, PCA, and kinship matrices, led to pronounced substructure within the population and enabled the capture of additional genomic loci, consistent with other studies (Altendorf et al., 2021; Beche et al., 2020), as well as identification of narrow genomic intervals in a locus‐dependent manner. As a result, the population was found to be suitable for mapping of quantitative, highly heritable agronomic traits such as DTH and PH.

By integrating whole genome sequence data, both generated from this study and publicly available sequences (Mansueto et al., 2017), this study facilitated QTL mapping with improved resolution that can enable direct identification of candidate genes within these regions demonstrating a major advance over conventional linkage mapping, which often relies on indirect inference. The rich genetic architecture and broad allelic diversity of the IRAT109 NAM population found in this study enabled GWAS to uncover both loci previously detected by conventional approaches and numerous unique QTLs, offering enhanced power to dissect the genetic basis of quantitative phenotypes.

The innovative value of this population is further underscored by its relevance to Australian rice breeding, where whole genome sequencing of varieties and genetic studies on complex traits like low‐temperature tolerance and aerobic adaptations have been notably limited. The diversity of the NAM parents ensured that the population captured a wide spectrum of alleles, maximizing the likelihood of uncovering novel genetic variation for these challenging traits. By rigorously validating QTL through candidate gene co‐localization and demonstrating the population's suitability for GWAS, this work provided both a practical toolkit and a precedent for future genetic dissection in japonica rice, with clear implications for breeding programs targeting stress resilience and productivity in temperate environments.

4.1. Sequence variation and phylogenetic analysis highlight the broad genetic diversity among the NAM parents

Evaluation of sequencing quality using whole‐genome data from the NAM parents revealed consistently high alignment rates across parents and acceptable Ti/Tv ratios, underscoring the robustness of mapping and minimal contamination (Pearson, 2013) and high‐quality SNP calls genome‐wide (DePristo et al., 2011; Y. Guo et al., 2012, 2013; Wang et al., 2014), closely aligning with published rice datasets (McNally et al., 2009; Subudhi et al., 2020). Such variation in sequencing depth and quality is a critical factor shaping variant detection and overall genomic data reliability (Cai et al., 2016; Goossens et al., 2021). Meanwhile, the variable percentages of properly mapped reads point to underlying differences in genomic complexity or potential technical artifacts (Dozmorov et al., 2015; Ross et al., 2013). These patterns are consistent with broader rice genomic studies, which highlight the role of structural variants and gene presence/absence variation in shaping genetic diversity (Shang et al., 2022).

Collectively, sequencing of the NAM parents demonstrated both technical robustness and genetic differences, which are central to understanding the genetic architecture of the NAM parents. The sequencing and variant‐calling metrics presented for the four parents (Langi, IRAT109, Norin PL8, and RL11) show sufficient coverage for confident genotyping. The resulting sequence data, as revealed by stringent alignment metrics and biologically relevant variant distributions, are valuable for downstream analyses of genetic structure.

Phylogenetic analysis results of NAM parents with a selected subset from the 3K RGP panel also suggested that NAM parents represent diverse genetic backgrounds with respect to a broader collection of rice genotypes. The IRAT109 NAM population parents were distributed across the japonica group, including both temperate and tropical japonica subpopulations, and not confined to a single clade. This broad distribution demonstrated that the NAM parents capture a wide spectrum of genetic diversity within the japonica group, which is critical for maximizing the utility of the IRAT109 NAM population in dissecting quantitative traits and enhancing breeding programs (Gireesh et al., 2021). By capturing a wide spectrum of genetic diversity within the parents, the resulting IRAT109 NAM population enables exploration of both common and rare variants, positioning it as a strategically important tool for future rice genomics.

4.2. Rich genetic architecture and diversity within the NAM population underpin its effectiveness for robust genetic analysis

The SNPs identified within the IRAT109 NAM population exhibited a nonuniform distribution, with certain chromosomal regions showing higher marker densities consistent with previous findings in rice and other cereal crops, where sequence diversity varied along chromosomes (Amadu et al., 2025; Nawade et al., 2024). Given a rice genome size of ∼430 Mb (Arumuganathan & Earle, 1991) and a genome‐wide LD decay distance of ∼1.6 Mb, the genome can be partitioned into approximately 260–270 independent LD blocks. The 4515 markers used in this study yield ∼16–17 markers per block, indicating dense coverage relative to the scale of LD in this population. The comprehensive genome coverage achieved in this study was critical for robust QTL mapping activities, maximizing the probability of detecting loci associated with quantitative traits.

PCA of the IRAT109 NAM population revealed pronounced population structure differentiating the three individual subpopulations. NAM parent lines (Langi, IRAT109, RL11, and Norin PL8) are clearly distinguishable and the common parent (IRAT109) is approximately mid‐way, demonstrating the expected relationship between half‐sib individuals in a NAM population (Altendorf et al., 2021). This pattern likely reflects the substantial genetic divergence among the three donor parents (Norin PL8, Langi, and RL11), which represent distinct temperate and tropical japonica backgrounds, providing the breadth of genetic diversity that underpins the utility of this population for quantitative trait dissection.

Kinship analysis reinforced these findings: higher degrees of relatedness within subpopulations and lower relatedness between subpopulations, which corroborate the PCA results. This high within‐family relatedness at the F6 generation suggests that donor‐specific haplotype blocks have been largely preserved through inbreeding, without residual segregation and genotyping artifacts. Such retention of intact haplotype blocks aligns with the NAM design principle of combining within‐family linkage power with the broad allelic diversity contributed by each donor parent (Fragoso et al., 2017; Yu et al., 2008).

Intersection analysis of SNP distribution highlights the genetic diversity contributed by each parental line and demonstrates the population's capacity to capture a wide spectrum of allelic variation. The enrichment of both common and rare alleles is a key advantage of the NAM design, facilitating the dissection of complex trait genetics (Gireesh et al., 2021; Yu et al., 2008). This advantage is reinforced by the faster LD decay observed in the IRAT109 NAM population, occurring up to 600 kb earlier than in the biparental populations, thereby enabling higher mapping resolution. Comparable patterns have been reported in durum wheat, where NAM populations also exhibited faster LD decay than their biparental population counterparts (Alahmad et al., 2023).

However, it is important to note that the four‐parent design of the IRAT109 NAM population is more limited in founder number than previous rice NAM studies that have employed 6, 10, and 11 parental lines (Fragoso et al., 2017; Kitony et al., 2021; Nawade et al., 2024). With only four founders, the allele frequency spectrum captured is inherently narrower, which can result in a reduced capacity to sample rare alleles across the japonica gene pool, lower detection power, and an inflated false discovery rate relative to larger NAM designs (Ogut et al., 2015; Yu et al., 2008). This limitation, however, reflects a deliberate trade‐off, prioritizing only complementary allelic variation for aerobic adaptation and low‐temperature tolerance within a genetically tractable japonica germplasm set relevant to Australian rice production.

Despite the restricted number of parents, the diversity captured across the parents proved sufficient to generate a genetically rich mapping population. Aerobic field phenotyping of the IRAT109 NAM population exhibited substantial phenotypic diversity for both DTH and PH across the years. The wide ranges in DTH and PH confirm the genetic potential of the NAM population to capture transgressive segregation, essential for dissecting the genetic control of these quantitative traits. Strong genotypic differences and high heritability values were evident, underscoring the NAM population's value for dissecting the genetic basis of quantitative traits.

The performance of the NAM population was benchmarked against the four NAM parents and three check varieties. Among the parents, Langi was the last to reach heading and IRAT109 was the tallest, while Norin PL8 was the earliest to reach heading and the shortest among parents. RL11 had an intermediate phenotype among the parents. The presence of parental lines with contrasting phenotypes enhances the mapping power of the panel, offering opportunities for identifying QTL linked to adaptive and agronomic traits under aerobic growing conditions. A similar approach has been done in an Ethiopian durum wheat NAM population (Kidane et al., 2019) where diverse parental lines were used for population development. Hung et al. (2012) also demonstrated that selecting phenotypically diverse parents is essential for developing highly informative and promising segregating populations.

Collectively, the results revealed comprehensive SNP coverage and clear population structure observed in the IRAT109 NAM population that provided a robust platform for the mapping of quantitative traits in rice. The genetic architecture and extensive phenotypic diversity, particularly for DTH and PH, ensure that this population will be invaluable for future genetic analyses, including GWAS, linkage mapping, and genomic selection, ultimately supporting rice improvement efforts, including its utility for elucidating the genetic basis of quantitative traits such as DTH and PH.

4.3. NAM GWAS uncovered unique loci undetected by biparental linkage mapping and enhanced locus resolution

Using both NAM GWAS and traditional biparental linkage mapping, a comprehensive set of genomic regions associated with DTH and PH was identified under aerobic field conditions. While DTH and PH are well‐established complex traits governed by numerous QTLs across diverse rice germplasm (Hori et al., 2016; Zhou et al., 2016), the genetic architecture detected in the present population is consistent with an oligogenic model. A total of 10 QTLs were detected in both NAM and biparental populations, indicating loci that contribute to trait variation across diverse analytical methods and genetic backgrounds. Additionally, NAM GWAS detected nine additional loci for DTH and PH, with the allelic effects accounted for by several NAM parents highlighting the enhanced power of IRAT109 NAM to dissect quantitative traits, attributable to broader allelic diversity present in the population (Gireesh et al., 2021). These unique loci underscore the value of NAM GWAS in uncovering previously undetected genetic variation, providing new opportunities to explore trait architecture and improve rice breeding strategies. Similar trends have been reported in wheatgrass (Altendorf et al., 2021) and soybean (Beche et al., 2020), where NAM populations revealed additional loci than biparental populations. It should be noted, however, that the larger combined sample size of the NAM population (n = 422) relative to individual biparental subpopulations (n = 106–174) may contribute to increased statistical power for QTL detection (Li et al., 2010). Nevertheless, several advantages, including faster LD decay, broader allelic coverage from multiple donor parents, and narrower QTL confidence intervals for several loci, are attributable to the NAM design structure rather than sample size alone.

Several loci were also exclusively detected in biparental populations. These QTLs likely reflect alleles that are confined to specific parental combinations (Wei & Xu, 2016) and therefore do not segregate broadly across the NAM parents. Additionally, some loci may have been excluded after accounting for population structure due to incorporation of PCs in the GWAS models (Yu et al., 2008). The absence in NAM may also reflect insufficient statistical power to detect rare QTL when dispersed across a more genetically diverse multi‐parent population where other families do not carry the significant QTL (Li et al., 2011). Similar results have been found by Ogut et al. (2015) using a maize NAM population where the authors argued that NAM can detect commonly segregating small‐effect QTL shared across families but lacks the power to detect strong but rare family‐specific QTL. Notably, the consistent detection of overlapping loci between biparental linkage mapping and NAM GWAS reinforces the biological relevance of these regions and builds confidence in the robustness of the associations.

In addition to broader genomic coverage, NAM GWAS also enabled locus‐dependent higher resolution identification of trait‐associated loci. For example, broad DTH loci on chromosomes 6, 7, and 12 and PH loci on chromosomes 1, 3, 7, and 9 were refined into substantially narrower intervals in the NAM population. The most striking improvement was the refinement of one genomic region from 6.88 Mb to just 0.54 Mb. However, two loci on chromosomes 6 and 7 identified in narrower intervals in biparental populations demonstrate that resolution gains from NAM are locus‐dependent rather than universal. Some regions benefit substantially from the accumulated recombination across multiple families, while others remain constrained regardless of population structure, as noted by McMullen et al. (2009). These results suggest that while NAM leverages on accumulated recombination across multiple families (Gireesh et al., 2021) that can lead to the discovery of narrower genomic intervals and finer scale dissection of signals, the magnitude of this resolution gain remains locus‐dependent. Nonetheless, these findings provide a demonstration of the power of IRAT109 NAM to resolve broad QTL intervals identified through biparental mapping in japonica rice, building on previous reports of QTL coincidence between NAM and biparental populations (Altendorf et al., 2021; Bajgain et al., 2016; Beche et al., 2020). These improvements are critical for downstream applications such as candidate gene identification, functional validation, and marker‐assisted selection. Although the current study did not perform parental projection‐based imputation, this approach could be applied in future work to further localize recombination breakpoints and refine QTL intervals (B. Guo & Beavis, 2011). The observed narrowing of intervals in the current study nevertheless reflects the resolution achievable with the present marker density in this IRAT109 NAM population.

It must be noted that within the NAM GWAS several loci were exclusively detected by either FarmCPU or BLINK but not both. This is attributable to fundamental differences in how these models iteratively select pseudo‐quantitative trait nucleotides as covariates to control for background genetic effects (Huang et al., 2018; Wang & Zhang, 2021). Because these covariate sets differ between models, the signal captured by each model may not be identical. This model‐specific detection behavior has been documented in recent GWAS in rice (Gong et al., 2025) and wheat (Milek et al., 2026), where FarmCPU and BLINK each identified unique marker‐trait associations not detected by the other. As such, loci detected by either GWAS model are considered valid candidates, while those detected by both represent the most robust associations.

Collectively, these findings demonstrate the complementary nature of biparental linkage mapping and NAM GWAS. While biparental linkage mapping offers robust detection of rare or background‐specific alleles that contribute to trait variation within individual families, NAM GWAS provides a more comprehensive and finer view of shared genomic regions controlling the genetic architecture underlying quantitative traits from multiple donors (Bouchet et al., 2017; Gireesh et al., 2021). Importantly, loci detected only in biparental populations but absent in NAM likely reflect allelic effects confined to a narrow genetic background. Ultimately, the regions shared between both approaches provide breeders with validated and reliable targets for marker‐assisted selection, while the novel loci discovered exclusively through NAM highlight additional genetic variation contributed by multiple parents for further investigation and crop improvement.

Beyond the analytical approaches employed here, joint QTL linkage mapping and whole‐genome nested association mapping (WGNAM) represent a complementary statistical framework for NAM populations. The joint inclusive composite interval mapping method proposed by Li et al. (2011) leverages variation in recombination frequencies across families by constructing a consensus linkage map from observed crossovers within each family, thereby exploiting the cumulative recombination information available across the full NAM population. Simulations showed improved mapping resolution over mapping using biparental populations; however, detection power drops considerably for QTL with small additive effects and joint mapping can obscure rare QTL signals, suggesting complementarity of approaches. WGNAM, as proposed by Paccapelo et al. (2022) extends this further by computing NAM founder probabilities, enabling the identification of QTL with contrasting allelic effects across founders. The underlying model is a linear mixed model that incorporates raw phenotype data, molecular markers, and population design. The authors note that the one‐stage approach improves the accuracy of estimated genetic effects, even for traits with low heritability. Future analyses of the IRAT109 NAM population using such frameworks could provide complementary linkage‐based validation of the loci identified here, as well as founder‐specific effect estimates enabled by the WGNAM approach, further resolving the contribution of individual parental alleles to trait variation.

4.4. Gene co‐localization confirms accuracy and robustness of NAM

The genomic regions highlighted in this study illustrate the reliability of the IRAT109 NAM population and support the biological plausibility of the detected QTL. By investigating the major‐effect QTL intervals (R ² > 10%) and leveraging parental sequence variation revealed through whole‐genome resequencing, the results are consistent with previously identified loci associated with DTH and PH. Specifically, two loci for DTH, qDTH6 and qDTH7.1, and one locus for PH, qPH7.1, colocalized with functionally characterized genes influencing heading time and PH.

qDTH7.1 and qPH7.1 are collocated in the same region in chromosome 7, suggesting possible shared genetic control between DTH and PH for the NAM population. Within this region, the donor allele associated with delayed heading consistently corresponded to increased PH, while allele conferring earlier heading was linked to reduced PH. Sequence variant analysis revealed that polymorphisms in Ghd7 co‐segregated with the donor alleles identified through GWAS, implicating this gene as a strong functional candidate gene for qDTH7.1 and qPH7.1. Ghd7 has been identified as a critical gene for flowering time regulation with pleiotropic effects on important agronomic traits such as culm length, panicle length and number, grain size, and head rice percentage (Lee et al., 2024; Sun et al., 2022; Zheng et al., 2019). It encodes a CONSTANS, CONSTANS‐LIKE, and TIMING OF CHLOROPHYLL A/B BINDING1 (CCT) domain protein that represses Early heading date 1 (Ehd1) under long‐day conditions, thereby suppressing florigen genes Hd3a and RFT1, delaying heading, and prolonging the vegetative stage resulting in increased PH (Itoh et al., 2010; Nemoto et al., 2016; Osugi et al., 2011; Xue et al., 2008). Under short‐day conditions, Ghd7 expression is downregulated, Hd3a and RFT1 expression peaks, and the plant transitions to the reproductive phase (Brambilla et al., 2017; Weng et al., 2014). Rising florigen levels subsequently downregulate PREMATURE INTERNODE ELONGATION 1 (PINE1), promoting internode elongation for panicle exertion at heading (Gómez‐Ariza et al., 2019). Consistent with this, Weng et al. (2014) revealed that elevated Ghd7 transcripts are quantitatively proportional to delayed heading and increased PH. Similarly, Lee et al. (2024) revealed that Ghd7 on PH is associated with over a 10 cm increase in culm length and accounts for up to 30% of culm length variation. The co‐localization of both qDTH7.1 and qPH7.1 with Ghd7, therefore, reflects a biologically coherent relationship, where an interconnected regulatory hub simultaneously shapes flowering time and plant architecture in rice.

On chromosome 6, qDTH6 colocalized with HEADING DATE 1 (HD1, LOC_Os06g16370), which promotes heading under short‐day conditions but suppresses it under long days (Yano et al., 2000). Also within the qDTH6 interval is SET DOMAIN GROUP 711 (SDG711, LOC_Os06g16390) that represses OsLF, which is a negative regulator of HD1 (Liu et al., 2014). Through this upstream action, SDG711 indirectly enhances HD1 expression, reinforcing the HD1‐mediated delay of flowering under long‐day conditions and promotion under short days, thereby increasing HD1 expression resulting in suppressed flowering under long‐day conditions and flowering under short days. Critically, SDG711 functions as a modulatory component within the HD1 regulatory pathway rather than an independent determinant of heading date. Thus, despite their co‐localization within qDTH6, the phenotypic variation captured by this QTL can be attributed to allelic variation at HD1, with SDG711 potentially contributing secondary regulatory influence through its upstream modulation of HD1 expression. Lastly, NITRATE TRANSPORTER 1/PEPTIDE TRANSPORTER 3.1 (OsNPF3.1, LOC_Os06g15370) colocalizes within this interval. However, the primary documented function of OsNPF3.1 has been associated with improved nitrogen use efficiency with pleiotropic effects in heading date, PH, and 1000‐grain weight (Yang et al., 2023). The influence of OsNPF3.1 on heading date, therefore, likely reflects an indirect response to variation in nitrogen status rather than a direct role in the flowering pathway. For this reason, OsNPF3.1 is considered the lowest priority candidate gene for qDTH6.

The co‐localization of mapped loci with known functional genes demonstrated that using GWAS in NAM identified biologically meaningful genetic regions. By anchoring significant SNPs to candidate genes with established roles in flowering and plant architecture, this study demonstrated that the loci detected were not merely statistical associations but are consistent with known regulators of DTH and PH, suggesting biologically plausible candidate genes that warrant functional validation. In addition to confirming known loci, NAM also revealed several promising regions for DTH (qDTH3.1) and PH (qPH1.2, qPH3.1, and qPH7.2) that may harbor novel genes governing these traits. The results of the current study emphasize the value of the IRAT109 NAM population as a powerful resource that enables the discovery of both established and novel determinants of key agronomic traits and facilitates the integration of whole‐genome resequencing data to strengthen genetic analyses.

5. CONCLUSION

The IRAT109 NAM population represents a powerful platform for dissecting the genetic architecture of quantitative agronomic traits in rice. By intercrossing multiple parents to generate RILs, NAM enables broad allelic sampling and recombination, offering greater mapping power and resolution than conventional biparental populations. Whole‐genome sequencing revealed extensive polymorphism among the four NAM parents, which span a wide breadth of japonica rice diversity, providing a robust foundation for genomic analysis.

SNP genotyping revealed clear population structure and substantial genetic variation across subpopulations. Combined with 2 years of field phenotyping, the population yielded highly heritable measurements for DTH and PH, with wide phenotypic ranges reflecting its capacity to generate transgressive segregants. Multi‐locus GWAS (FarmCPU and BLINK) identified robust associations for both traits, refining broad QTL intervals into narrower genomic regions at specific loci, while biparental linkage mapping captured family‐specific loci reflecting rare allelic effects. Many detected loci colocalized with well‐characterized genes underlying these traits, underscoring the complementary value of both approaches.

Beyond DTH and PH, the diverse genetic backgrounds of the parental lines position the IRAT109 NAM population as a promising resource for studying low‐temperature stress resilience in rice. Future work will apply this population to key developmental stages under low‐temperature growing conditions. The four‐parent design does limit the breadth of allelic diversity captured, highlighting opportunities for population expansion or complementary mapping strategies to fully resolve the genetic basis of quantitative traits.

AUTHOR CONTRIBUTIONS

Fergie Ann Quilloy: Conceptualization; data curation; formal analysis; funding acquisition; methodology; visualization; writing—original draft; writing—review and editing. Ricky Vinarao: Conceptualization; methodology; supervision; writing—review and editing. Christopher Proud: Conceptualization; formal analysis; methodology; supervision; writing—review and editing. Thi Quynh Chang Nguyen: Methodology; writing—review and editing. Dara Daygon: Supervision; writing—review and editing. Shu Fukai: Supervision; writing—review and editing. Jaquie Mitchell: Conceptualization; funding acquisition; supervision; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Quilloy, F. A. , Vinarao, R. , Proud, C. , Nguyen, T. Q. C. , Daygon, D. , Fukai, S. , & Mitchell, J. (2026). An IRAT109 NAM population: Genetic characterization and mapping utility for trait dissection in japonica rice (Oryza sativa L.). The Plant Genome, 19, e70309. 10.1002/tpg2.70309

DATA AVAILABILITY STATEMENT

Raw phenotype data, genotype data, and GWAS results are available upon request.