Genome-wide association analysis identifies natural allelic variants associated with leaf rolling in rice (Oryza sativa L.)

Abstract

Background and Aims

Leaf rolling in rice affects plant architecture, light interception and photosynthetic efficiency. This study aims to identify the genetic basis of adaxial leaf rolling using a genome-wide association study (GWAS) in a diverse panel of rice accessions.

Methods

A GWAS was performed using 1.5 million single nucleotide polymorphisms (SNPs) across 113 rice accessions. The analysis included population structure, linkage disequilibrium, haplotype patterns and expression correlation. To validate the identified association, an additional panel of japonica inbred lines was analyzed, supporting the link between the SNP and adaxial leaf rolling.

Key Results

A japonica-specific quantitative trait locus, qALR1, was identified on chromosome 1. A putative causal SNP was located within a shared cis-regulatory motif (RAV1-B) upstream of two candidate genes, LOC_Os01g72370 and LOC_Os01g72380. Previous studies suggest that variation in this motif can influence gene expression, supporting the hypothesis that regulatory divergence in these genes may underlie variation in leaf rolling. Validation with an expanded panel of japonica inbred lines revealed its enrichment in breeding germplasm, indicating possible historical selection.

Conclusions

This study identifies a subspecies-specific regulatory mechanism underlying adaxial leaf rolling in rice. The putative causal SNP may affect transcription factor binding and gene regulation, ultimately influencing leaf morphology. These findings offer valuable targets for optimizing canopy architecture and enhancing photosynthetic performance through molecular breeding.

INTRODUCTION

Leaves are the major organs of plants that function as solar panels, in which photosynthesis converts carbon dioxide and water into carbohydrates and oxygen, respectively. Leaf shape, which is determined by parameters such as leaf length, width, area, thickness and degree of curling, is an important factor that influences photosynthetic capacity (Tessmer et al., 2013; Zhang et al., 2021; Huang et al., 2022). Generally, broader and flatter leaves have higher maximum photosynthetic rates (Stenberg et al., 1995). However, these leaf shapes do not always correlate positively with enhanced canopy photosynthetic efficiency. As the leaves become wider and flatter, the upper leaves tend to shade the lower leaves, eventually decreasing the overall photosynthetic capacity of the plant (Duncan, 1971; Stewart et al., 2003; Alves et al., 2024).

For nearly three decades, farmers and breeders have targeted the V-shaped or moderately curled leaf type as a key trait in rice breeding (Wu 2009; Yuan, 2017). This is because moderate leaf rolling stiffens the leaf blades and maintains leaf erectness (Hernandez and Brebbia, 2012). This minimizes mutual shading among leaves and enhances light penetration, thereby improving the photosynthetic rate (Zhu et al., 2001; Yang and Hwa, 2008). Additionally, under stress conditions, such as drought, high light intensity and high temperature, moderate leaf rolling increases stomatal resistance and maintains high humidity near the leaf surface, reducing water loss (Kadioglu and Terzi, 2007). However, severe leaf rolling can lead to delayed plant growth, developmental defects and reduced yields (Yang et al., 2014; Effendi et al., 2019; Wang et al., 2023). Therefore, developing plant types with moderate leaf rolling would enhance photosynthetic efficiency and yield optimization, which continues to be the central focus of rice breeding programmes (Setter et al., 1995).

Advancements in molecular methodologies have led to the discovery of more than 60 leaf rolling genes, which have been cloned and functionally characterized in rice. Fu et al. (2019) categorized these genes into five groups based on their roles in the final appearance of leaves: leaf polarity, bulliform cells, sclerenchyma cells, commissural veins and cuticle development. Among the three axial polarities, the adaxial–abaxial polarity is known to have the greatest influence on leaf morphology (Jiajia et al., 2020). Genes such as OsAGO7, OsYABBY1, OsHox32/OsHB4, OsAGO1a and OsAGO1b have been identified to regulate leaf rolling by controlling the development of different tissues on the adaxial and abaxial sides (Dai et al., 2007; Shi et al., 2007; Itoh et al., 2008; Li et al., 2013, 2016, 2019). Bulliform cells are highly vacuolated cells that are located in the vasculature of the adaxial leaf surface. Defective development, which alters the number, size and distribution of bulliform cells, affects the degree of leaf rolling. Most cloned leaf rolling genes, such as OsACL1, OsLC2/OsVIL3, OsRL14, OsSRL1/CLD1, OsREL1, OsROC5 and OsYUC8/OsNAL7/OsCOW1, belong to this group (Li et al., 2010; Zhao et al., 2010; Fang et al., 2012; Xiang et al., 2012; Chen et al., 2015; Xu et al., 2021; Zhou et al., 2023). Sclerenchyma cells, which are supporting tissues usually composed of dead cells, also play an important role in the maintenance of flat leaves. Deficiencies of OsSLL1/RL9, OsSRL2/OsAVB/OsNRL2 and OsSND2, whose function is to determine the development of sclerenchyma cells in the leaf, lead to defects in sclerenchyma cell formation, resulting in a leaf rolling phenotype (Zhang et al., 2009; Liu et al., 2016; Ye et al., 2018). OsCVD1 regulates the development of commissural veins that transversely connect the adjacent longitudinal veins. Jing et al. (2017) hypothesized that a deficiency in forming the commissural veins in cvd1 mutants weakens the connection between longitudinal veins, leading to the observed leaf rolling phenotype. OsCFL1 belongs to the fifth group of genes that regulate leaf rolling by participating in cuticle development. Both the cfl1 mutant and overexpressed OsCFL1 transgenic plants showed impaired cuticle development with a leaf rolling phenotype (Wu et al., 2011).

Although the identification of various leaf rolling genes has improved our understanding of the molecular basis of leaf rolling, the application of these genes to breeding programmes has been challenging. This is because most of the related mutants exhibit severe leaf rolling, along with other abnormal morphological traits. Therefore, efforts have been made to identify natural sequence variations in key leaf rolling quantitative trait loci (QTLs)/genes (Zhang et al., 2016; Jang et al., 2021; Meng et al., 2021; Chen et al., 2023). However, many of the genes associated with leaf rolling remain to be fully elucidated. Consequently, the identification of leaf rolling genes and large-effect alleles from the natural rice population will facilitate future targeted genetic modifications to improve plant architecture, photosynthetic efficiency and, ultimately, rice yield.

In this study, we investigated the genetic mechanisms regulating adaxial leaf rolling in 113 rice accessions. A genome-wide association study (GWAS) for adaxial leaf rolling identified loci associated with the trait and pinpointed genomic regions harbouring candidate genes and major alleles. We further examined how allelic variation at these loci contributes to natural variation in the leaf rolling index (LRI). The findings of this study will contribute to our understanding of the genetic basis underlying the leaf rolling trait in rice.

MATERIALS AND METHODS

Plant materials and growing conditions

In this study, two panels of rice accessions obtained from the National Agrobiodiversity Center (NAC) of the Rural Development Administration (RDA) in Korea were used for both phenotypic evaluation and genetic analysis. The first panel, an association panel consisting of 113 rice accessions exhibiting varying degrees of adaxial leaf rolling, was selected from the Korean World Rice Collection (KRICE) and used for QTL mapping by GWAS. Details of this association panel have been previously described by Kim et al. (2016) and Aung et al. (2023). The second panel, composed of 118 japonica inbred accessions developed in Korea, was used to assess whether the single nucleotide polymorphism (SNP) identified through GWAS exhibited a consistent effect across additional japonica accessions. This allowed for further validation of the genetic influence of the identified SNP.

A total of 113 accessions from the association panel and 118 japonica inbred accessions were planted for phenotypic measurements in a field at Pusan National University, Miryang, South Korea, in 2022 under irrigated conditions. Plants were sown at the beginning of March and transplanted 30 d after sowing into the paddy field as follows: a single accession per row, 12 plants per row, 15 cm distance between plants in a row and 30 cm distance between adjacent rows.

Phenotypic measurements and statistical analysis

The LRI was measured 2 weeks after heading, when the neck of the panicle had completely emerged from the leaf sheath. To prevent the acceleration of leaf rolling caused by environmental factors such as drought stress, strict water management was implemented, and the paddy field was continuously irrigated until the LRI measurements were completed to minimize the influence of soil water content on leaf rolling. Three replicates of LRI per accession were measured from the centre of the flag leaf, and the average was used for further analysis. As the LRI exhibited significant deviations from the normal distribution (Supplementary Data Fig. S1), the Box and Cox transformation method was employed to identify the optimal transformation for LRI (Box and Cox, 1964). The LRI was calculated using the following equation (Shi et al., 2007):

$$\begin{array}{rl}& \mathrm{LRI}(\text{\%})\\ & ={\displaystyle \frac{(\mathrm{Width}\mathrm{of}\mathrm{fully}\mathrm{expanded}\mathrm{leaf}\mathrm{blade}\text{–}\mathrm{Natural}\mathrm{distance}\mathrm{of}\mathrm{leaf}\mathrm{blade}\mathrm{margin})}{\mathrm{Width}\mathrm{of}\mathrm{fully}\mathrm{expanded}\mathrm{leaf}\mathrm{blade}}}\\ & \times 100\end{array}$$

Genotypic data

Genomic sequencing data for KRICE were obtained from previously reported studies (Kim et al., 2016; Aung et al., 2023), which were generated using an Illumina HiSeq 2500 sequencing platform. The resequencing data were processed in a multistep pipeline involving data preparation, filtering, mapping, sorting, and variant calling procedures. VCFtools (v.0.1.15) were used to remove missing values and heterozygotes from the raw data (Danecek et al., 2011). These filtered high-quality reads were aligned to the Nipponbare genome sequence (IRGSP 1.0) to compare the output sequences among the accessions using SAMtools v.1.3.1 (Li et al., 2009). Duplicate reads aligned at multiple locations were removed using PICARD v.1.88. Variant calling was then performed using GATK tools v.4.2.6.1 (McKenna et al., 2010).

To examine the distribution of the putative causal SNP identified in this study within a broader rice population, we utilized the SNP-Seek database (https://snp-seek.irri.org/index.zul) to assess its allele composition across the International Rice Research Institute (IRRI)’s 3K panel. This allowed for a comparative analysis of the SNP’s presence in diverse rice accessions and provided insights into its subspecies specificity.

Population structure and genetic relationships

The population structure was initially analysed using principal component analysis (PCA) based on two principal components (PC1 and PC2) with the R package Genomic Association and Prediction Integrated Tool (GAPIT) (Lipka et al., 2012). The results were visualized using the ggplot2 package in R. To further investigate population structure, ADMIXTURE (v.1.3.0) software (Alexander et al., 2009) was used with K values ranging from 1 to 10, and the optimal K value was determined using the Evanno method by calculating ΔK (Evanno et al., 2005). Additionally, phylogenetic analysis was conducted using IQ-TREE v.2.3.4 based on 1.5 million high-quality SNPs, with 1000 bootstrap replicates. The best-fit model, TVM+F+I+R8, was identified using ModelFinder and implemented in IQ-TREE (Nguyen et al., 2015). Linkage disequilibrium (LD) was calculated with SNPs among the 113 rice accessions using PopLDdecay software (v.3.40) and the squared correlation coefficient (r²) (Zhang et al., 2019). LD data were summarized by estimating the mean LD between a pair of SNPs and plotted against the physical distance using the MeanBin method to visualize LD decay.

Genome-wide association study

We conducted a GWAS to identify genomic regions associated with the adaxial leaf rolling trait using 1.5 million high-quality SNP markers. The analysis was performed in R with the GAPIT package (v.3.0) using the Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway (BLINK) model (Wang and Zhang, 2021). To address the issue of multiple testing, the Bonferroni correction was applied based on SNP markers, resulting in a genome-wide significance threshold of 3.31 × 10⁻⁸ [−log₁₀(P) ≈ 7.48]. Manhattan plots for all associations and their corresponding quantile–quantile (QQ) plots were drawn using the ggplot2 package in R.

Furthermore, in silico analysis was conducted to identify genomic regions harbouring the SNP–trait associations and to explore potential candidate genes underlying the observed genotype–phenotype relationships. The genomic region of the QTL was defined based on the average LD decay distance of the corresponding chromosome, extending both upstream and downstream of the lead SNP identified through the association analysis. Additionally, LDBlockShow software was used to generate LD heatmaps for visualization (Dong et al., 2021).

Identification and haplotype analysis of candidate genes

To identify potentially novel candidate genes associated with adaxial leaf rolling, protein sequences of previously reported leaf rolling-related genes in rice, obtained from published studies on leaf rolling, were retrieved from the Rice Annotation Project Database (https://rice.uga.edu/index.shtml, accessed in January 2025). These sequences were then used as queries to perform protein–protein BLAST (BLASTP) searches against the NCBI protein database (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE=Proteins, accessed in January 2025) to determine whether any of the candidate genes identified within the QTL region corresponded to these previously reported genes.

For the candidate genes located within the identified QTL region, haplotype analysis was performed by using the multcompView package in R. Haplotypes were inferred based on SNPs located in the promoter region (2 kb upstream of the ATG start codon), non-synonymous SNPs within the coding sequence (CDS), splice-site SNPs and variants located in the untranslated region (UTR). Only haplotypes with a frequency of ≥5 % among accessions retained after quality control (removal of heterozygous accessions) were considered for further analysis. Phenotypic differences among haplotypes were evaluated using one-way analysis of variance (ANOVA), followed by Tukey’s honest significant difference (HSD) test for pairwise comparisons.

To assess the cumulative effect of multiple candidate genes within the QTL region and to capture potential gene–gene interactions, combined haplotype analysis was conducted by integrating haplotypes derived from these genes (Kõks et al., 2004). The combined haplotypes were subjected to the same statistical comparison using ANOVA and Tukey’s HSD test. Additionally, to investigate haplotype diversity and infer evolutionary relationships among haplotypes, a haplotype network analysis was performed using the Median Joining Network method implemented in PopART software (v.1.7) (Leigh and Bryant, 2015).

Functional and regulatory annotation of putative causal SNPs

To compare the amino acid sequences associated with the two major haplotypes (high and low LRI haplotypes), multiple sequence alignments were conducted, incorporating the reference sequence from Nipponbare. In addition, to determine whether the putative causal SNPs identified in this study were located within known functional elements, the corresponding amino acid sequences were analysed using the SMART database (http://smart.embl-heidelberg.de/, accessed in January 2025) to assess their positions relative to annotated protein domains.

Similarly, to investigate whether the putative causal SNPs identified in the promoter regions were located within known cis-acting regulatory elements relevant to plant adaptation and signal transduction, promoter sequences of the candidate genes were retrieved from Oryzabase (https://shigen.nig.ac.jp/rice/oryzabase/about/oryzabase, accessed in January 2025) and analysed using the New PLACE database (https://www.dna.affrc.go.jp/PLACE/?action=newplace, accessed in January 2025). The results were visualized using MEME Suite v.5.5.7 (https://meme-suite.org/meme/tools/meme).

Expression correlation analysis

To explore the regulatory relationships between candidate genes identified in this study and previously reported leaf rolling-related genes, an expression correlation analysis was performed using publicly available transcriptome data from the Rice Oligonucleotide Array Database v.2 (ROADv2, https://roadv2.khu.ac.kr/rice/correlation/). A total of 60 known leaf rolling-related genes were selected to assess their correlation with GWAS-identified candidate genes. Pearson correlation coefficients (r) were calculated, and only significant correlations (P < 0.05) with r > 0.5 were considered for further analysis. This criterion ensured that only strong and biologically relevant associations were included in the interpretation of gene expression patterns. The Nipponbare cultivar used to construct ROADv2 carries an allele at the putative causal SNP that is commonly found in both japonica and indica accessions, rather than being japonica-specific (Cao et al., 2012; Hwang et al., 2024).

RESULTS

Variation of adaxial leaf rolling trait

We used an association panel of 113 rice accessions grown in 2022 to investigate the phenotypic variation in adaxial leaf rolling. The adaxial LRI of the flag leaf was assessed 2 weeks after heading, when the panicle was fully emerged from the leaf sheath. Since the phenotypic distribution was skewed toward a low LRI (Supplementary Data Fig. S1), we applied the Box–Cox transformation method (Ntakirutimana et al., 2023) to normalize the data. After transformation, the LRI distribution closely approximated a normal distribution (Fig. 1A). Information on the mean, range, standard deviation and coefficient of variation (CV) of the LRI is provided in Table S1. A wide range of phenotypic variability was observed among the 113 accessions, with a CV of 52.61 %. Among the subspecies, the temperate japonica group exhibited the highest average LRI (1.85), while the admixture group had the lowest average LRI (1.43). However, no significant differences were detected among the subspecies (Fig. 1B). On the other hand, when comparing LRI between japonica (including temperate japonica and tropical japonica) and indica (including aus, admixture and indica), a significant difference was observed between the two groups (Fig. 1C). Additionally, the CV was higher in indica than in japonica (Table S1).

Fig. 1.

Phenotypic variation and population structure analysis. (A) Histogram of the adaxial LRI distribution across all accessions after Box–Cox transformation. (B) Box plot comparing LRI among subspecies. Groups include temperate japonica (TEJ), tropical japonica (TRJ), aus (AUS), admixture (ADMIX) and indica (IND). Different letters indicate significant differences among groups (P < 0.05, Tukey’s HSD test). (C) Box plot comparing LRI between japonica and indica groups. An asterisk (*) indicates a statistically significant difference based on Student’s t-test (P < 0.05). (D) PCA of the rice accessions. The plot illustrates genetic differentiation among subspecies along PC1 and PC2. (E) Population structure analysis using ADMIXTURE. The plot shows ΔK values across different cluster numbers (K), with K = 2 identified as the most likely population structure. (F) Population structure assignment at K = 2. The plot highlights genetic differentiation between japonica (red) and indica (blue) accessions. (G) Phylogenetic tree among rice accessions. Japonica accessions are shown in red and indica accessions in blue to depict genetic relationships.

Variant identification, population structure and linkage disequilibrium

In the present study, 113 rice accessions exhibiting adaxial leaf rolling were selected from the previously fully sequenced KRICE population (Kim et al., 2016; Aung et al., 2023) for variant identification. After read filtering with MAF > 5 %, missing data < 20 % and proportion of heterozygous variants < 5 %, we obtained 1 509 860 high-quality SNPs, with an average of 4.3 variants per kb. Using these SNP markers, we first analysed the population structure of 113 rice accessions, comprising 42 temperate japonica, 17 tropical japonica, six aus, 18 admixture and 30 indica, through PCA based on two principal components. The PCA results revealed that the first two PCs primarily distinguished the japonica and indica groups, with PC1 and PC2 explaining 55.09 and 5.38 % of the total genetic variance, respectively (Fig. 1D). Furthermore, population structure and phylogenetic tree analyses were performed. Based on the ΔK values calculated, the most appropriate number of clusters was K = 2, primarily separating japonica and indica, which is consistent with the PCA results (Fig. 1E–G). Taken together, our population structure analysis indicates that the genotypes of this association panel depict a strongly structured population. Therefore, these results were incorporated as covariates in the subsequent analyses.

The average genome-wide LD decay distance, at which r² dropped to half of its maximum value (Kim et al., 2007), was ∼220, 150 and 220 kb in the whole, japonica and indica association panels, respectively (Supplementary Data Fig. S2). The LD decay rate for each chromosome was estimated using the same approach. The LD decay rate varied considerably among different chromosomes and exhibited distinct patterns across different populations. In the whole association panel, the LD decay rate for all chromosomes, except chromosomes 1, 2 and 3, reached half of its maximum value at distances ranging from approximately 15 kb (Chr 11) to 133 kb (Chr 7) between the SNP markers (Figure S2A). However, chromosomes 1, 2 and 3 exhibited strong LD patterns extending beyond 200 kb. In the japonica association panel, the LD decay rate for all chromosomes, except for chromosomes 1, 3 and 5, varied from ∼11 kb (Chr 11) to 78 kb, while chromosomes 1, 3 and 5 displayed strong LD patterns extending beyond 120 kb (Fig. S2B). In the indica association panel, the average LD decay distances were generally shorter than those observed in the japonica panel (Fig. S2C).

Detection of QTLs associated with leaf rolling by GWAS

A set of 1 509 860 high-quality SNPs, uniformly distributed across all 12 chromosomes (Supplementary Data Fig. S3), was used to identify associations between SNPs and the adaxial leaf rolling trait in 113 rice accessions. Given that both the LRI and population structure exhibited significant differences between the japonica and indica association panels, steps were taken to minimize potential biases caused by population structure and to gain a more precise understanding of the genetic basis underlying the adaxial leaf rolling trait. To achieve this, a GWAS was conducted not only on the whole panel but also separately for each subspecies.

Using the BLINK method, a single identical association signal was detected on chromosome 1 in both the whole association panel and the japonica panel (Fig. 2A and B), while no significant association was identified in the indica panel (Fig. 2C). The QQ plots showed a uniform distribution of observed P-values, with slight inflation in the whole and japonica panels, but not in the indica panel (Fig. 2D–F). This pattern suggests that the GWAS model was appropriately fitted to the data and that potential genetic signals exist specifically in the whole and japonica panels. These findings collectively indicate that the identified association signal is derived primarily from the japonica subspecies, while the indica subspecies probably possesses a distinct genetic background for the adaxial leaf rolling trait.

Fig. 2.

GWAS analysis of adaxial leaf rolling in rice. (A–C) Manhattan plots showing the GWAS results for (A) the whole panel, (B) japonica panel and (C) indica panel. The x-axis represents the genomic position across the 12 rice chromosomes, and the y-axis shows the –log₁₀(P) values of SNP associations. The horizontal dashed line indicates the genome-wide significance threshold. (D–F) QQ plots corresponding to the GWAS results for (D) the whole panel, (E) japonica panel and (F) indica panel.

The genomic region of the QTL was defined based on the average LD decay distance of chromosome 1 in the japonica panel, which was calculated to be ∼158 kb. To ensure sufficient coverage of the associated region, we extended the interval to 200 kb upstream and downstream of the lead SNP identified through the association analysis. This region, spanning 41.78–42.18 Mb, was designated as adaxial leaf rolling 1 (qALR1) (Fig. 3 and Supplementary Data Table S2). Next, we assessed whether the QTL identified in this study overlapped with previously reported QTLs associated with the leaf rolling trait from other GWAS and QTL studies. While no QTLs directly overlapped with the QTL region identified in this study, a QTL for leaf rolling on chromosome 1, previously identified in a doubled-haploid rice population under drought stress conditions (Courtois et al., 2000), was located in close proximity to qALR1 (Table S2).

Fig. 3.

Genomic landscape of qALR1. The top panel presents the local Manhattan plot for the qALR1 region, showing SNP association signals with adaxial leaf rolling. The horizontal red line indicates the genome-wide significance threshold. The middle panel illustrates the distribution of candidate genes within qALR1, where yellow, red, black, light blue and grey boxes represent expressed genes, hypothetical proteins, transposon-related proteins, candidate genes and other genes within the candidate region, respectively. The bottom panel displays the LD heatmap, where red intensity represents stronger LD (higher r² values), providing insight into the LD structure surrounding the candidate region. Black triangles indicate LD blocks.

Identification of candidate genes and haplotypes associated with leaf rolling trait

Since extensive LD can encompass a large number of candidate genes, making downstream analysis more challenging, we applied local pairwise LD correlation (r² > 0.6) around the lead SNP to define a more refined candidate genomic region and reduce the number of candidate genes (Ntakirutimana et al., 2023). This analysis revealed that distinct LD blocks were formed around the lead SNP in the japonica panel, whereas no clear LD block was observed in either the whole population or the indica panel (Fig. 3 and Supplementary Data Fig. S4). Based on the japonica-specific LD structure, we selected the genes located within this LD block as candidate genes associated with the adaxial leaf rolling trait and proceeded with further analyses.

Within the qALR1 region, we identified six distinct LD blocks, each containing more than 10 SNPs. Notably, the lead SNP was located in the first LD block, which harboured a total of 34 candidate genes, excluding retrotransposons and transposon-related proteins (Fig. 3 and Supplementary Data Table S3). To date, more than 60 genes associated with the leaf rolling trait have been reported in rice (Table S4). To assess the potential colocalization between these known genes and the candidate region, we examined whether any of these previously identified genes overlapped with qALR1. However, no known leaf rolling genes colocalized with this region, suggesting that qALR1 may harbour novel genetic factors underlying the adaxial leaf rolling trait that have not been previously reported.

To identify novel candidate genes associated with adaxial leaf rolling, we employed two strategies. The first involved using NCBI’s BLASTP function to identify homologues of the 60 known leaf rolling-related genes and determine whether any of the candidate genes corresponded to them. A total of 45 homologues were distributed across all 12 chromosomes, with six located on chromosome 1 (Supplementary Data Table S4). However, none of the candidate genes matched these homologues. The second strategy involved haplotype analysis using SNPs located within the intragenic region as well as the 2-kb upstream region from the start codon. Among the 34 candidate genes, four (LOC_Os01g72240, LOC_Os01g72310, LOC_Os01g72370 and LOC_Os01g72380) were identified to have distinct haplotypes, which exhibited significant differences in their effects on the adaxial leaf rolling index (Fig. 4 and Fig. S5).

Fig. 4.

Haplotype structure and phenotypic effects of four candidate genes. (A–D) Haplotype analysis of (A) LOC_Os01g72240, (B) LOC_Os01g72310, (C) LOC_Os01g72370 and (D) LOC_Os01g72380. Each diagram illustrates the genomic structure of the respective gene, with haplotype alignments showing sequence variations. Sequence differences among haplotypes are indicated by coloured cells, and a putative causal variant is highlighted in red. While a total of 101 SNPs were identified in this region, only three representative SNPs are linked to the gene structure diagram and haplotype alignments. (E–H) Phenotypic comparison of haplotypes for (E) LOC_Os01g72240, (F) LOC_Os01g72310, (G) LOC_Os01g72370 and (H) LOC_Os01g72380. The box plots display LRI variation among different haplotypes, with significant differences indicated by different letters (P < 0.05, Tukey’s HSD test). Each dot represents an individual accession. (I–L) Haplotype network analysis for (I) LOC_Os01g72240, (J) LOC_Os01g72310, (K) LOC_Os01g72370 and (L) LOC_Os01g72380. The size of each circle represents the number of accessions carrying a particular haplotype, while colours indicate subspecies composition, distinguishing temperate japonica (TEJ) and tropical japonica (TRJ) accessions.

LOC_Os01g72240 was mapped 89.27 kb upstream of the lead SNP within the LD block (Fig. 3). Ten sequence variations were identified, including four SNPs in the promoter region, one in the 5′-UTR, three in the 3′-UTR, and one non-synonymous SNP (Fig. 4A). Based on these variations, three haplotypes were identified (Fig. 4E). Among them, HAP1 and HAP2 exhibited significant differences in adaxial LRI values, with HAP1 accessions displaying higher values than HAP2. Both haplotypes were exclusively found in temperate japonica rice accessions (Fig. 4I). LOC_Os01g72310 was mapped 54.82 kb upstream of the lead SNP within the LD block (Fig. 3). Seven sequence variations were identified, including one SNP in the promoter region, three non-synonymous SNPs and three in the 3′-UTR (Fig. 4B). Based on these variations, three haplotypes were identified (Fig. 4F). Among them, HAP1 and HAP2 showed significant differences in adaxial LRI values, with HAP2 accessions exhibiting higher values than HAP1. Both haplotypes were exclusively found in temperate japonica rice accessions (Fig. 4J). Additionally, LOC_Os01g72370 and LOC_Os01g72380 flanked the lead SNP within the LD block, positioned 1.71 kb upstream and 1.99 kb downstream, respectively (Fig. 3). LOC_Os01g72370 exhibited 10 sequence variations, including five SNPs in the promoter region, one in the 5′-UTR and four in the 3′-UTR (Fig. 4C). Based on these variations, three haplotypes were identified, with HAP1 and HAP3 exhibiting significantly higher adaxial LRI values than HAP2 (Fig. 4G). HAP1 and HAP2 were predominantly or exclusively composed of temperate japonica rice accessions, with HAP1 also including some tropical japonica accessions. In contrast, HAP3 consisted primarily of tropical japonica accessions, with a minor proportion of temperate japonica accessions (Fig. 4K). LOC_Os01g72380 exhibited two sequence variations, both located in the promoter region (Fig. 4D). Three haplotypes were identified, with HAP1 and HAP2 showing significantly higher adaxial LRI values than HAP3 (Fig. 4H). Similar to LOC_Os01g72370, HAP1 encompassed all subspecies, with temperate japonica as the predominant type, whereas HAP2 contained both subspecies in reverse proportions. In contrast, HAP3 was exclusively found in temperate japonica rice accessions (Fig. 4L). These findings suggest that genetic variations in LOC_Os01g72240, LOC_Os01g72310, LOC_Os01g72370 and LOC_Os01g72380 play a crucial role in regulating adaxial LRI, with distinct haplotypes contributing to phenotypic differences observed exclusively in japonica rice accessions.

Combined haplotype analysis of four candidate genes

To evaluate the cumulative effect of the four candidate genes, capture potential gene–gene interactions and identify the gene most likely to be responsible for the observed phenotypic differences, we performed a combined haplotype analysis (Fig. 5A). This analysis targeted the entire 99.97-kb region encompassing the four candidate genes. Within this region, including both intragenic and promoter regions but excluding intergenic regions, a total of 101 sequence variations were identified. Based on these variations, five haplotypes were defined, each shared by at least five accessions. HAP1–HAP4 exhibited highly similar genotypes, differing at only three positions. While HAP1–HAP3 showed comparable adaxial LRI values, HAP4 displayed a significantly lower LRI value compared to the other haplotypes (Fig. 5B). This phenotypic difference was associated with an SNP at position 41 980 126, where the allele changed from A to C. Notably, this SNP is a shared variant between LOC_Os01g72370 and LOC_Os01g72380, as LOC_Os01g72370 is transcribed in the reverse direction and LOC_Os01g72380 in the forward direction, resulting in overlapping promoter regions due to their close proximity (∼3.7 kb apart). All HAP1–HAP4 haplotypes were exclusively composed of temperate japonica rice accessions (Fig. 5C). In contrast, HAP5, which exhibited a distinct genotype compared to the other haplotypes, consisted primarily of tropical japonica accessions, with some temperate japonica accessions. These findings suggest that the SNP at position 41 980 126, located within the common promoter region of LOC_Os01g72370 and LOC_Os01g72380, is a putative japonica-specific causal SNP contributing to natural variation in adaxial LRI.

Fig. 5.

Combined haplotype analysis and functional characterization of the candidate genomic region. (A) Haplotype analysis across a genomic region containing four candidate genes. Sequence polymorphisms are indicated by coloured cells, and key variants, including a putative causal SNP, are highlighted in red. (B) Phenotypic comparison of haplotypes. The box plot illustrates LRI variation among different haplotypes, with significant differences denoted by different letters (P < 0.05, Tukey’s HSD test). Each dot represents an individual accession. (C) Haplotype network analysis across temperate japonica (TEJ) and tropical japonica (TRJ) accessions. The size of each circle represents the number of accessions carrying a given haplotype, and colours indicate subspecies composition. (D) Identification of cis-regulatory elements in the promoter regions of candidate genes. The line, inverted triangles and coloured boxes (grey, pink, red, blue and green) represent the promoter, SNPs, intragenic regions, motif 1 (RAV1-A), motif 2 (RAV1-B), motif 3 (GTGANTG) and motif 4 (EBOXBNNAPA/MYCCONSENSUSAT), respectively. (E) Sequence logo of the RAV1-B motif showing nucleotide variation that may affect transcription factor binding.

Functional analysis of SNPs in candidate genes

To assess the potential function of the common SNP in the promoter region of LOC_Os01g72370 and LOC_Os01g72380, we examined whether the SNP associated with phenotypic differences between haplotypes was positioned within known promoter motifs. Since LOC_Os01g72370 encodes a basic helix–loop–helix (bHLH) DNA-binding domain-containing protein, which plays a key role in plant adaptation and signal transduction (Gao and Dubos, 2024), we identified four relevant motifs in the promoter regions of both genes using New PLACE: RAV1-A, RAV1-B, GTGANTG and EBOXBNNAPA/MYCCONSENSUSAT (Fig. 5D). In total, seven and four SNPs were identified in the promoter regions of LOC_Os01g72370 and LOC_Os01g72380, respectively. However, a single SNP, located at the same position in both gene’s promoters, was found within the RAV1-B motif (Fig. 5E). Given that the RAV1-B motif is recognized by the RAV1 (Related to ABI3/VP1 1) transcription factor, which regulates gene expression and is involved in various physiological and developmental processes in plants (Matías-Hernández et al., 2014), this shared SNP may serve as a key regulatory element, influencing the expression of both genes and affecting the degree of leaf rolling.

Allelic variation and potential selection in a japonica inbred panel

To further investigate whether this putative japonica-specific putative causal SNP within the promoter region of LOC_Os01g72370 and LOC_Os01g72380 also influences adaxial leaf rolling across diverse japonica accessions, we analysed allelic variation and LRI in a set of 177 japonica accessions. This set included the japonica GWAS panel and an additional 118 japonica inbred accessions, comprising both temperate and tropical japonica. A total of 12 SNPs were identified in the genomic region, including the intragenic and promoter regions of both LOC_Os01g72370 and LOC_Os01g72380. These included six SNPs in the common promoter region, one in the 5′-UTR and five in the 3′-UTR of LOC_Os01g72370 (Fig. 6A). Based on these variations, four haplotypes were identified, each shared by at least 17 accessions. Consistent with findings from the japonica GWAS panel, HAP2, which carries the A-to-C allele change at position 41 980 126, exhibited a significantly lower adaxial LRI value compared to other haplotypes (Fig. 6B). With the exception of HAP3, all haplotypes were primarily or entirely composed of temperate japonica accessions (Fig. 6C). Similarly, HAP2, which showed a significantly lower adaxial LRI, was predominantly found in temperate japonica accessions, rather than tropical japonica.

Fig. 6.

Haplotype analysis and network of the candidate genomic region using the GWAS panel and additional japonica inbred lines. (A) Haplotype alignment of the genomic region containing the putative causal SNP within the shared promoter of two divergently transcribed genes. The diagram at the top illustrates the genomic structure, with arrows indicating gene orientations. Sequence variations are colour-coded, and key SNPs are highlighted in red. (B) Phenotypic comparison of haplotypes. The box plot presents significant differences among haplotypes, denoted by different letters (P < 0.05, Tukey’s HSD test). Each dot represents an individual accession. (C) Haplotype network analysis. Circle size corresponds to the number of accessions per haplotype, and colours represent subspecies composition in temperate japonica (TEJ) and tropical japonica (TRJ) accessions.

Interestingly, in the japonica inbred panel, the proportion of accessions carrying the C allele at position 41 980 126 was 34.7 % (41 out of 118) (Fig. 6A), an increase of 25.9 % compared to the japonica GWAS panel, which showed 8.8 % (five out of 57), as well as the 3K panel from the IRRI, which showed 6.0 % (50 out of 828) (Supplementary Data Table S5). These findings suggest that this japonica-specific SNP may have been subject to selection pressure and that japonica accessions with substantial leaf rolling may have been selectively bred toward moderate leaf rolling, possibly as an adaptive trait in breeding programmes.

DISCUSSION

Leaves are the primary organs for light capture and play a crucial role in photosynthesis, with their morphology being directly linked to photosynthetic efficiency. In rice, leaf rolling is a key determinant of leaf shape, influencing light interception and canopy structure. In this study, we identified significant genetic variation in adaxial leaf rolling among 113 rice accessions representing diverse genetic backgrounds. GWAS analysis revealed the genetic basis of leaf rolling variation between japonica and indica panels, identifying a major japonica-specific QTL associated with this trait.

Leaf rolling differences between japonica and indica rice

Population structure analysis, including PCA, admixture analysis and phylogenetic tree construction, revealed that the 113 accessions could be broadly classified into two major groups: japonica (including temperate and tropical japonica) and indica (including aus, admixture and indica) (Fig. 1D–G). Notably, japonica exhibited a significantly higher LRI compared to indica (Fig. 1C), aligning with the general tendency of japonica rice to exhibit a more erect plant architecture (Gong et al., 2014; Wei et al., 2020). It has been suggested that increased leaf rolling in japonica may contribute to greater leaf stiffness, which in turn supports a more upright growth habit (Hernandez and Brebbia, 2012). However, the causal relationship between leaf rolling, leaf stiffness and plant architecture remains to be fully established. While previous studies indicate that leaf rolling enhances leaf stiffness, it is unclear whether this relationship is consistent across japonica and indica rice. Additional mechanical stiffness assays and gene expression analyses are needed to confirm this link, as other factors, including leaf geometry, growth patterns and hormonal regulation, may also contribute to plant architecture.

Identification of a subspecies-specific QTL for adaxial leaf rolling

Given the distinct phenotypic and genetic differences between japonica and indica, we conducted GWAS for adaxial leaf rolling in both the entire population and the subpopulations separately. A single significant SNP was consistently detected in both the whole panel and the japonica panel (Fig. 2). However, despite the high CV for adaxial LRI in indica (Supplementary Data Table S1), which suggests substantial genetic variation, no significant association signals, including the japonica-identified SNP, were detected in indica. This result suggests that the identified SNP may have a subspecies-specific effect, which could be explained by multiple factors. One possibility is that the causal variant near this locus is absent in the indica panel. Alternatively, even if the variant is present, it may either be entirely unrelated to the trait in indica while being associated in japonica, or fail to meet the stringent significance threshold required for GWAS, possibly due to the reduced statistical power following population subdivision.

These findings suggest that genetic regulation of leaf rolling differs between japonica and indica, with subspecies-specific variants playing a crucial role. This highlights the importance of considering population structure when interpreting GWAS results and developing functional markers for breeding applications. Additionally, differences in the genetic architecture of adaxial leaf rolling between japonica and indica suggest that this variant may have a more significant role in japonica than in indica.

Discovery of candidate genes and functional insights into adaxial leaf rolling

To further explore the genetic basis of adaxial leaf rolling, we defined a QTL region (qALR1) surrounding the lead SNP, extending 200 kb upstream and downstream (Fig. 3). Within this region, LD block analysis identified six LD blocks, with the first block containing the lead SNP. Among the 34 candidate genes within this block (excluding transposon-related proteins), none overlapped with the previously reported 60 leaf rolling-related genes or their homologues. This suggests the presence of a novel, previously uncharacterized gene associated with leaf rolling.

Haplotype analysis further refined the candidate gene list, revealing that only four genes exhibited significant LRI differences between haplotypes (Fig. 4). Among them, a putative causal SNP was identified between LOC_Os01g72370 and LOC_Os01g72380 as having the strongest effect on adaxial leaf rolling (Fig. 5A and B), suggesting that the SNP may regulate the expression of both genes, influencing the observed phenotypic differences. Notably, haplotype analysis of the whole panel (Supplementary Data Fig. S5C and D) revealed that this SNP was absent in the indica panel, potentially explaining the lack of a significant GWAS signal in indica. Furthermore, an evaluation of the 3K rice panel maintained by IRRI confirmed that this SNP is also absent in indica accessions, reinforcing its japonica-specific nature.

To assess the presence of this japonica-specific putative causal SNP in other japonica populations, we conducted an additional haplotype analysis using genomic data from japonica inbred lines. Interestingly, the frequency of the haplotype carrying the allele associated with a lower LRI was significantly higher in these inbred lines compared to the japonica GWAS panel and the 3K panel (Fig. 6A and B, and Supplementary Data Table S5). Initially found at a low frequency in the japonica panel, this allele appears to have increased in japonica inbred lines, probably as a result of breeding selection. These findings suggest that although japonica rice generally exhibits a higher degree of leaf rolling than indica, selective breeding may have influenced this trait. Specifically, breeders may have favoured lines with moderate rolling (lower LRI) over time, leading to a gradual decrease in the overall LRI of japonica inbred lines cultivated in Korea.

These findings highlight the role of artificial selection in shaping leaf rolling traits, in addition to natural genetic variation. The exclusive presence of this SNP in japonica, along with its altered haplotype distribution in breeding lines, underscores the importance of subspecies differentiation and selection pressure in the genetic regulation of leaf morphology.

Regulatory mechanisms and functional implications of candidate genes

To investigate the potential regulatory mechanism of this SNP, we analysed cis-regulatory elements within the promoter regions of LOC_Os01g72370 and LOC_Os01g72380. Notably, the SNP was located within a RAV1-B motif, a putative binding site for B3 domain-containing transcription factors (Fig. 5D and E). These transcription factors, including RAV1, play critical roles in plant development by regulating gene expression through direct promoter binding. They are involved in stress responses, hormone signalling and organ morphogenesis (Hu et al., 2004; Woo et al., 2010; Feng et al., 2014; Fu et al., 2014; Mandal et al., 2023). The RAV1-B motif, characterized by the conserved CACCTG sequence, has been identified as a key regulatory element for transcription factor binding (Kagaya et al., 1999).

Notably, the functional significance of SNP-induced changes in cis-regulatory motifs has been demonstrated in various systems. For example, a study by Tsujimura et al. (1996) showed that in the c-kit promoter, both the deletion of a CACCTG motif-containing sequence and a single SNP mutation (CTCCAG substitution) disrupted MITF-mediated transcriptional activation. Although this example originates from a mammalian system, it underscores the potential impact of single-nucleotide changes within regulatory motifs on transcription factor binding and gene expression. Similarly, it is possible that the putative causal SNP identified in this study alters RAV1-B motif function, affecting transcriptional regulation of LOC_Os01g72370 and LOC_Os01g72380, and consequently influencing adaxial leaf rolling.

Additionally, LOC_Os01g72370 encodes a bHLH transcription factor, which is generally involved in developmental processes and environmental responses (Gao and Dubos, 2024). Several bHLH transcription factors have been implicated in leaf rolling regulation in tomato, further supporting the potential role of this gene in rice (Zhu et al., 2017; Li et al., 2022; Guo et al., 2023). In contrast, LOC_Os01g72380, annotated as an expressed protein, lacks direct functional characterization but was co-expressed with the highest number of known leaf rolling-related genes (21 out of 60, 35 %) (Supplementary Data Table S6). This suggests that LOC_Os01g72380 may influence leaf rolling through coordinated gene expression rather than a direct regulatory role.

Based on these findings, we hypothesize that the putative causal SNP within the RAV1-B motif influences the transcriptional activity of LOC_Os01g72370 and LOC_Os01g72380, either directly or indirectly modulating leaf rolling by altering transcription factor binding affinity. Further studies, including electrophoretic mobility shift assays (EMSA) and chromatin immunoprecipitation (ChIP), are necessary to validate whether this SNP affects transcription factor binding and gene regulation. Additionally, functional validation using transgenic plants will be essential to confirm its role in regulating adaxial leaf rolling.

CONCLUSION

This study provides new insights into the genetic regulation of adaxial leaf rolling in rice. Our GWAS analysis identified a japonica-specific QTL, with a putative causal SNP located within a RAV1-B motif, a recognized binding site for B3 domain-containing transcription factors. LD and haplotype analyses further identified LOC_Os01g72370 and LOC_Os01g72380 as strong candidate genes that may be regulated by this SNP. The potential cis-regulatory mechanism suggests that this SNP modulates transcription factor binding affinity, thereby affecting gene expression and leaf rolling phenotype. LOC_Os01g72370, encoding a bHLH transcription factor, is functionally linked to plant morphological traits, including leaf rolling, while LOC_Os01g72380 exhibits strong co-expression with known leaf rolling-related genes, suggesting an indirect role in leaf morphology regulation. Additionally, haplotype analysis in japonica inbred lines revealed that the frequency of the allele at this SNP, which is associated with a lower LRI, increased significantly in breeding lines. This suggests that selection pressure may have influenced leaf rolling traits over time, potentially favouring moderate-rolling phenotypes in japonica cultivars. Despite these findings, further experimental validation is required to confirm the direct impact of this SNP on transcriptional regulation and leaf rolling. Future studies should focus on functional assays, such as EMSA and ChIP, as well as transgenic approaches, to establish the causal relationship between this SNP and adaxial leaf rolling. These insights will deepen our understanding of leaf morphology regulation and may contribute to breeding strategies aimed at optimizing light interception and photosynthetic efficiency in rice.

Funding

This work was supported by a grant from the Ministry of Science and ICT (MSIT), Republic of Korea [RS-2022-00165400], as well as by grants from the Rural Development Administration (RDA), Republic of Korea [RS-2022-RD010353].

Supplementary data

Supplementary data are available at Annals of Botany  online and consist of the following. Figure S1: Histogram showing the skewed distribution of the LRI. Figure S2: LD decay in 113 rice accessions. The genome-wide average LD decay pattern is shown as the relationship between r² values and the physical distance between SNP pairs in (A) the whole panel, (B) the japonica panel and (C) the indica panel. Figure S3: SNP distribution and density along the 12 chromosomes. Colours correspond to the number of SNPs in a 1-Mb region. Figure S4: Local Manhattan plots and LD heatmaps for (A) the whole panel and (B) the indica panel. The corresponding plots for the japonica subspecies are included in the main body of the paper. Figure S5: Haplotype structure and phenotypic effects of four candidate genes in the using the whole association panel. (A–D) Haplotype analysis of four candidate genes. (E–H) Phenotypic comparison of haplotypes based on the LRI. (I–L) Haplotype network analysis for each candidate gene.

Author contributions

D.L. was involved in conducting the experiments, analysing the data, interpreting the results and drafting the manuscript. A.R.L. and J.G. were involved in the analysis and interpretation of phenotypic data; Y.Z. and B.Y.J. were involved in the analysis and interpretation; B.K., S.G.J. and S.W.K. were involved in revising the manuscript; J.L. conceived the study and was involved in the critical revision of the manuscript and the final approval of the version to be published.