GWAS and Candidate Gene Prediction of Elemental Accumulation Traits in Rice Using a Multiparental Population

Abstract

Rice plants accumulate essential elements to sustain physiological processes during growth and development and to ensure the nutritional quality of the grain as a food source. However, the genetic basis of elemental accumulation and the interrelationships among elemental concentrations across different tissues remain poorly understood. To conduct breeding aimed at improving the absorption characteristics of multiple interrelated elements, genetic analysis using experimental populations that retain diversity while sharing the genetic background of cultivated varieties is effective. Here we show genetic variations in the concentrations of 13 elements (P, K, Ca, Mg, As, Cd, Cr, Cu, Fe, Mo, Mn, Ni, and Zn) in rice straw at the flowering stage and grain at the mature stage using a multi-parent advanced generation inter-cross (MAGIC) population that derived from eight cultivars including both Japonica and Indica. Comprehensive evaluation of the correlation coefficients revealed divergences in the association between grain and straw for several combinations of elements. Haplotype-based genome-wide association studies (GWAS) identified 51 and 53 quantitative trait loci (QTLs) in straw and grain, respectively. In total, the 104 QTLs were grouped into 19 clusters and 60 independent QTLs. By leveraging the haplotype information from the MAGIC population, 52 candidate genes associated with the accumulation of Ca, Mg, Cd, Cu, Fe, and Mo were efficiently predicted from these QTLs, including both previously reported and novel genes. Among them, OsMOT1;1 encoding a molybdenum transporter, was predicted to be within a QTL associated with Mo accumulation in grain on chromosome 8. OsACA9, a homolog of autoinhibited Ca²⁺-ATPases, was predicted within a QTL related to Ca accumulation in straw on chromosome 2. In addition, an unidentified gene, OsCML6, which is presumed to be involved in calcium signaling, was predicted to be a candidate for a Ca-accumulation QTL on chromosome 11. These findings offer insights into haplotypes and putative genes associated with element accumulation and trait interrelationships, providing valuable information for optimizing plant growth and enhancing grain nutritional quality in rice breeding programs.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12284-026-00908-6.

Background

Rice (Oryza sativa L.) is one of the most important and well-established crops, serving as a staple food for nearly half of the world’s population (Khush 2005). The success of the “Green Revolution” in the mid-20^th century was a landmark event that allowed the world to see the utility of scientific breeding (Evenson and Gollin 2003; Pingali 2012). However, recent climate changes and fluctuations in global agricultural markets have generated numerous new concerns, necessitating the development of rice varieties with diverse and improved traits. One such concern is inefficient elemental uptake and imbalances in soil nutrient utilization associated with soil degradation, which often limit rice productivity (Ye et al. 2024).

Soil contains elements that are essential for plant growth and can be categorized based on plant requirements. Macroelements, which are required in large amounts for plant metabolism and growth, include C, N, P, K, Ca, Mg, and S (Hawkesford 2012). Trace elements, which are required in minimal quantities, account for 5 to 200 ppm, or less than 0.02% dry weight of plants and include Fe, Mn, B, Zn, Cu, Cl, Ni and Mo (Kathpalia 2018). Some of these elements are basic components of the plant protoplasm and components or activators of enzymes, whereas others can regulate protoplasmic membrane permeability, participate in buffer systems, and maintain cell permeability to sustain normal life activities (Taiz 2015). Lack of these elements causes physiological imbalances, affects growth and development, and manifests as specific symptoms of deficiency. In addition, both beneficial and toxic elements are present in soil. Beneficial elements can promote the growth and development of certain plants. For example, Si enhances the growth of melon and vegetables, thereby improving their resistance (Ma and Yamaji 2006). Toxic elements, such as Cd and As, not only impede plant growth but are also harmful to humans (Uchimiya et al. 2020). Beyond their importance for plant physiology, the elemental composition of rice grains directly affects human nutrition and food safety. Extensive datasets linking variations in grain element concentrations with potential dietary relevance are available in food composition databases (USDA FoodData Central, https://fdc.nal.usda.gov/). Food safety standards also specify limits for toxic elements (Codex 2025). Genetic variation in rice elements significantly influences both aspects. Therefore, elucidating the physiological roles of mineral elements and the genetic basis of their uptake and transport in plants can inform optimized nutrient management and breeding strategies to enhance yield and improve crop quality.

There is no doubt that crop genomic information, which is being rapidly revealed, makes an important contribution to breeding. After the rice genome was decoded (IRGSP 2005), the isolation and functional analysis of a large number of genes have advanced notably (Chen et al. 2022). This information has not only contributed to plant sciences but has also set a new trend in rice breeding. In particular, DNA marker-assisted breeding has greatly simplified the process of evaluating traits, which has been a challenge in conventional breeding, by increasing selection efficiency and reducing selection times (Yamamoto et al. 2009). However, the genomic history of rice breeding has revealed a narrow genetic diversity of current rice varieties, which will become a potential genetic resource for future breeding (Yonemaru et al. 2012). To ensure sustainable genetic improvement, new breeding techniques are required to maximize the potential diversity offered by combinations of breeding lines.

To expand genetic diversity by collapsing the genome structure in a crossbreeding population, a multiparent advanced generation inter-cross (MAGIC) population, consisting of multiple crosses derived from three or more founder varieties, has been proposed (Cavanagh et al. 2008; Kover et al. 2009). Since MAGIC populations have the advantage of QTL detection accuracy compared to single-cross populations and can produce gene combinations as diverse as outcrossing populations, they have been developed in many crop species such as tomatoes (Pascual et al. 2015), cowpea (Huynh et al. 2018), maize (Dell’Acqua et al. 2015), cotton (Islam et al. 2016), wheat (Huang et al. 2012; Mackay et al. 2014), barley (Sannemann et al. 2015), eggplant (Mangino et al. 2022), and rice (Bandillo et al. 2013; Meng et al. 2015; Ogawa et al. 2018a). The multifounder design in the MAGIC population ensures that each parent contributes to at least one target trait and that several parents may harbor favorable alleles for the same trait. This strategy greatly increases the diversity and combinations of functional alleles, thereby substantially enriching genetic diversity. Wang P. et al. (2024) demonstrated that the MAGIC population has great potential for exploring novel alleles of days to heading in natural or breeding rice varieties.

Most genetic analyses of elemental accumulation in rice have been conducted using mutants and distantly related varieties exhibiting extreme phenotypes. Extensive genetic and physiological studies comparing the functions of these alleles with those of the currently cultivated varieties have elucidated the molecular mechanisms underlying element uptake and translocation. However, to fully utilize this genetic information in practical breeding, it is necessary to understand the variation in the number and types of genes underlying these genetic mechanisms within the natural variations of rice currently used in crossbreeding programs. In this context, comprehensive genetic analysis using MAGIC, which comprises diverse varieties within a target region, is considered a highly effective approach. The elemental contents of plant tissues are thought to be subject to natural selection within the minimum acceptable levels necessary for survival. In other words, as this trait was not emphasized until very recently in modern rice breeding, it is expected that many alleles contributing to the fine-tuning of elemental content are hidden within breeding populations.

We created a MAGIC population from multiple crosses of eight high-yielding rice varieties in Japan (Ogawa et al. 2018a). Novel phenotypes that circumvent the tradeoffs between grain length and width were identified in this population (Ogawa et al. 2018b). This feature is unique compared to conventional single-cross populations. In this study, we conducted a genome-wide association study (GWAS) on the concentrations of 13 elements in rice grain and straw using this MAGIC population. This study has two objectives. The first is to utilize the unique characteristics of the MAGIC population to elucidate the genetic basis of variation in elemental concentration in rice. The second is to promote the effective exploration of candidate genes that affect this trait in breeding populations.

Materials and Methods

Materials and Field Test

A scheme for the development of the 8-way MAGIC population (Ogawa et al. 2018a) is shown in Supplementary Fig.S1. Eight founders representing high-yield varieties in Japan–Hokuriku 193 (HO), Mizuhochikara (MI), Suweon 258 (SU), Akidawara (AK), Takanari (TK), Bekogonomi (BE), Ruriaoba (RU), and Tachiaoba (TC)–were selected as founders of the MAGIC population. Four two-way hybrids (AB, CD, EF, and GH) were generated, followed by two four-way hybrids (ABCD and EFGH) and two eight-way F₁ hybrids (ABCDEFGH). The 182 lines of the MAGIC population (F₁₀) and their eight founders were planted in 2019 (sown on May 20, transplanted on June 20) and 2021 (sown on May 19, transplanted on June 19) at the experimental field of the Institute of Plant Science and Resources, Okayama University (34° 35’ 31’’ N and 133° 46’ 7’’ E), Kurashiki, Japan. Plants were grown in two adjacent plots within the same field; straw samples were collected from one plot, and grain samples were collected from the other plot. In each plot, six plants were planted in each of the four rows, resulting in a total of 24 plants. To determine the appropriate time for sampling, the days to heading were recorded for all lines. Heading was defined as the stage at which the panicle of the highest culm emerged at least 2 cm from the flag-leaf sheath.

Measurement of Element Contents

The straw at the flowering stage and grain at the mature stage were sampled and subjected to elemental analysis. After drying, the grain was hulled to obtain brown rice. The straw was dried at 70 °C for 24 h, and ground into a fine powder. Brown rice and dried powder (around 0.2 g) of the straw were taken and digested with nitric acid at a concentration of 61% for 10 to 15 h in a thermostatic metal bath (Dry Thermo Unit) at a temperature of up to 135 °C. The samples were filled with ultrapure water prior to analysis. The concentrations of 13 elements (P, K, Ca, Mg, As, Cd, Cr, Cu, Fe, Mo, Mn, Ni, and Zn) were determined using inductively coupled plasma mass spectrometry (ICP-MS; 7700X, Agilent Technologies, Santa Clara, CA, USA) in the helium mode.

Data Visualization

The correlation coefficient matrix heatmap was generated using Chiplot (Ji et al. 2022; Li et al. 2023), and the frequency distribution plot and box plot were produced using the “ggplot2” package in R (Wickham 2010). Manhattan plots were drawn using the “CMplot” package (Yin et al. 2021). QTL genetic maps derived from the MAGIC population were constructed using MapChart version 2.32 (Voorrips 2002).

Statistical Analysis

All statistical analyses were performed using the R software version 4.2.3 (R Core Team 2018). Standard functions such as cor() and aov() from the built-in stats package were used to conduct correlation analysis and one-way ANOVA, respectively, with the significance threshold set at p < 0.05. The quantitative data of the elements across different years were standardized using z-score normalization prior to model fitting. The statistical significance of differences among the means was assessed using the Tukey–Kramer multiple comparison test.

GWAS and QTL Definition

Haplotype data corresponding to 13,603 SNPs in the MAGIC population have been reported by Ogawa et al. (2018a). Haplotype prediction followed the method used for the Arabidopsis MAGIC population (Kover et al. 2009). Associations between the imputed genotype and phenotype data for all 13 elements in the MAGIC population were analyzed using the Kruskal–Wallis test to calculate p-values. P-values were obtained for the sample types of grain and straw in 2021 (Table 1) using GWAS. To control for false positives, p-values were adjusted using the Benjamini–Hochberg (BH) method implemented in the p.adjust() function, and the resulting adjusted p-values were interpreted as false discovery rate (FDR)-controlled q-values. The experiment was conducted with pre-set thresholds of FDR ≤ 0.1 and FDR ≤ 0.05, and QTL screening was performed accordingly. QTLs were named using the following convention: “q” followed by the last two digits of the year (21), sample type (G for grain, S for straw), elemental symbols, and chromosome numbers (e.g., q21GCd12 and q21SMg4).

Table 1.

Summary of significant QTLs (FDR ≤ 0.1) involving elemental concentrations in rice grain and straw using the MAGIC population

Grain Element	QTL name	SNP marker	–log10(p)	FDR	Straw Element	QTL name	SNP marker	–log10(p)	FDR
	q21GCd3-1	Chr3_27703252	2.93	0.0453		q21SCa1-1	Chr1_30341513	2.62	0.0718
	q21GCd3-2	Chr3_30422353	3.57	0.0209		q21SCa1-2	Chr1_35886973	2.34	0.0906
	q21GCd4-1	Chr4_1015760	2.42	0.0916		q21SCa2-1	Chr2_1003070	4.38	0.0358
	q21GCd4-2	Chr4_4598416	2.46	0.0879		q21SCa2-2	Chr2_9936760	3.05	0.0423
	q21GCd5	Chr5_24799047	2.38	0.0944		q21SCa2-3	Chr2_12082859	2.67	0.0718
Cd	q21GCd6-1	Chr6_818002	2.93	0.0453	Ca	q21SCa2-4	Chr2_28817978	2.82	0.0553
	q21GCd6-2	Chr6_1947565	4.58	0.0191		q21SCa6-1	Chr6_906853	3.44	0.0358
	q21GCd6-3	Chr6_26383319	2.46	0.0879		q21SCa6-2	Chr6_27609600	3.75	0.0358
	q21GCd7-1	Chr7_5852706	4.38	0.0191		q21SCa7	Chr7_25637166	2.77	0.0618
	q21GCd7-2	Chr7_21290846	2.37	0.0944		q21SCa10	Chr10_772038	2.31	0.0942
	q21GCd9	Chr9_13860903	2.64	0.0700		q21SCa11	Chr11_20019091	3.97	0.0358
	q21GCd11-1	Chr11_7024776	3.23	0.0263		q21SMg1	Chr1_38968519	2.90	0.0646
	q21GCd11-2	Chr11_21722693	2.41	0.0916		q21SMg2	Chr2_27493441	2.07	0.0924
	q21GCd11-3	Chr11_24809829	2.52	0.0791		q21SMg3-1	Chr3_8182099	2.02	0.0968
	q21GCd12	Chr12_3481562	2.38	0.0944		q21SMg3-2	Chr3_30422353	2.34	0.0769
Cu	q21GCu1-1	Chr1_3465892	2.81	0.0682		q21SMg3-3	Chr3_33913703	3.61	0.0646
	q21GCu1-2	Chr1_30397482	3.39	0.0610		q21SMg4	Chr4_23235680	2.75	0.0646
	q21GCu2	Chr2_23652007	3.01	0.0610	Mg	q21SMg5-1	Chr5_903828	2.94	0.0646
	q21GCu4-1	Chr4_5005293	2.92	0.0649		q21SMg5-2	Chr5_20100264	2.21	0.0824
	q21GCu4-2	Chr4_32654876	2.43	0.0917		q21SMg6-1	Chr6_2898917	2.03	0.0965
	q21GCu6-1	Chr6_272094	2.80	0.0682		q21SMg6-2	Chr6_25127567	2.87	0.0646
	q21GCu6-2	Chr6_23133577	2.56	0.0805		q21SMg7-1	Chr7_4139442	3.33	0.0646
	q21GCu7	Chr7_5775203	3.58	0.0610		q21SMg7-2	Chr7_23879876	2.68	0.0664
	q21GCu11	Chr11_19450668	3.04	0.0610		q21SMg8-1	Chr8_1152627	2.44	0.0762
	q21GFe1	Chr1_1928663	2.17	0.0983		q21SMg8-2	Chr8_27896450	2.29	0.0786
	q21GFe2	Chr2_22089977	2.17	0.0983		q21SMg9-1	Chr9_16610541	2.05	0.0942
	q21GFe3	Chr3_211120	2.50	0.0807		q21SMg9-2	Chr9_21306262	2.94	0.0646
	q21GFe4-1	Chr4_11760	4.09	0.0634		q21SMg10-1	Chr10_87974	2.32	0.0776
	q21GFe4-2	Chr4_7380693	2.24	0.0974		q21SMg10-2	Chr10_19811172	2.37	0.0762
	q21GFe4-3	Chr4_8125665	2.86	0.0634		q21SMg11-1	Chr11_1293419	3.51	0.0646
	q21GFe4-4	Chr4_16598461	3.37	0.0634		q21SMg11-2	Chr11_7666770	2.07	0.0931
Fe	q21GFe6-1	Chr6_2898917	2.16	0.0983		q21SMn1	Chr1_22617559	2.77	0.0973
	q21GFe6-2	Chr6_11885263	2.93	0.0634		q21SMn2-1	Chr2_9619	2.99	0.0973
	q21GFe6-3	Chr6_16804756	2.17	0.0983		q21SMn2-2	Chr2_3370339	3.72	0.0768
	q21GFe6-4	Chr6_18841514	2.37	0.0891		q21SMn4	Chr4_19629130	3.51	0.0973
	q21GFe6-5	Chr6_26742584	2.26	0.0974	Mn	q21SMn5	Chr5_27586439	4.52	0.0444
	q21GFe7-1	Chr7_3468635	2.60	0.0737		q21SMn6-1	Chr6_3322080	2.64	0.0995
	q21GFe7-2	Chr7_5208158	3.18	0.0634		q21SMn6-2	Chr6_27061467	2.88	0.0973
	q21GFe7-3	Chr7_14863454	2.25	0.0974		q21SMn11-1	Chr11_3788078	2.95	0.0973
	q21GFe7-4	Chr7_27803788	2.31	0.0919		q21SMn11-2	Chr11_19284207	3.03	0.0973
	q21GFe8	Chr8_3960499	2.12	0.0995		q21SMo1	Chr1_28965344	6.39	0.0005
	q21GFe12-1	Chr12_4687453	3.48	0.0634		q21SMo2-1	Chr2_3964230	2.40	0.0827
	q21GFe12-2	Chr12_10465894	3.09	0.0634		q21SMo2-2	Chr2_16095083	2.30	0.0966
Mn	q21GMn5	Chr5_25805065	3.82	0.0670		q21SMo2-3	Chr2_22180762	2.39	0.0838
	q21GMo1	Chr1_26290272	2.74	0.0530		q21SMo3	Chr3_211120	2.29	0.0974
	q21GMo2	Chr2_35261376	2.43	0.0936	Mo	q21SMo4	Chr4_7150241	2.31	0.0959
	q21GMo3	Chr3_23153229	2.40	0.0969		q21SMo6-1	Chr6_139521	3.33	0.0207
	q21GMo6	Chr6_26742584	2.54	0.0785		q21SMo6-2	Chr6_3541729	2.35	0.0912
Mo	q21GMo7-1	Chr7_4926246	5.36	0.0050		q21SMo9	Chr9_22419659	2.28	0.0986
	q21GMo7-2	Chr7_19693968	3.12	0.0266		q21SMo10	Chr10_18064070	3.57	0.0164
	q21GMo8	Chr8_72552	3.60	0.0132		q21SMo11	Chr11_23755723	3.60	0.0164
	q21GMo12-1	Chr12_126100	2.58	0.0727
	q21GMo12-2	Chr12_18278484	3.70	0.0117

QTLs shown in bold italic indicate an FDR ≤ 0.05

Candidate Gene Estimation by Haplotype Comparison

Candidate gene estimation was performed near the QTL regions identified by the GWAS. After setting the threshold as noted in the previous section, all SNP loci were considered “significant” in this study when FDR ≤ 0.1. We conducted keyword searches in the Oryzabase (Yamazaki et al. 2010) and RAP-DB (Sakai et al. 2013) databases using three types of keywords and commonly studied agronomic traits. The element-related keywords consisted of the element name alone and in combination with the terms “ion” and “transporter” (e.g., “cadmium,” “cadmium ion,” and “cadmium transporter”). In addition, searches were performed using trait-related terms such as “heading date”, “tiller number”, and “detoxification”. Gene information related to keywords, including OsID, was downloaded and merged into a single file. Furthermore, genes located within QTL peak regions were also considered candidates even without keyword-based annotations if they satisfied criteria: (1) they exhibited significant phenotypic differences among haplotype groups based on multiple comparisons, and (2) their genotype patterns were identical to or similar to those of nearby screened genes. Additionally, for the screened genes, information such as gene location, functional description, or publication of the OsID was retrieved from the RAP-DB. Polymorphic information of the screened genes among the eight founders was obtained using TASUKE+ (Kumagai et al. 2019). By sorting the polymorphisms in the genetic information, the haplotypes of the eight founders were classified into several groups. Mean trait values were calculated for each of the eight founder alleles and each classified haplotype group, and then multiple comparisons using Tukey–Kramer tests were performed to confirm the relationship between the haplotype groups and phenotypic values. If a significant difference was observed between groups, the gene was considered a candidate gene.

Results

General Profiles of Element Concentration in Straw and Grain

We compared the trends of 13 element concentrations (P, K, Ca, Mg, As, Cd, Cr, Cu, Fe, Mo, Mn, Ni, and Zn) in grain measured in 2019 and 2021 (Supplementary Tables S1, S2, and S3). We observed a positive correlation between the years except for Cr in grain (Supplementary Fig. S2 A and B), indicating that environmental differences across the years did not substantially affect the relative accumulation patterns. This consistency indicates that the MAGIC population is suitable for the genetic analysis of grain element concentrations. The present study focused on conducting further analyses of the 2021 data set.

Descriptive statistics of the 13 element concentrations in both the MAGIC population and founders are summarized. Mean elemental concentrations differed markedly between grain and straw: most elements were higher in straw than in grain, whereas Mg and P were slightly higher in grain, and Cd was strongly enriched in grain (Supplementary Tables S4 and S5). Overall, the MAGIC lines exhibited wider value ranges across most elements than the founders. To further evaluate phenotypic variability, coefficients of variation (CV) were calculated. In grain, the CV of most elements was higher in the MAGIC population than in the founders (Supplementary Table S4), indicating increased variability in the MAGIC population. In contrast, other elements (P, Mn, Fe, Ni, and Mo) showed slightly higher CV values in the founder group. Similar patterns were also observed in straw (Supplementary Table S5). In addition to the previously mentioned elements, P, Mn, and Fe exhibited noticeably higher CVs in the MAGIC population. These results demonstrate a general expansion of phenotypic variation in the MAGIC population compared to the founders. The distributions showed continuous variation with transgressive segregants, indicating that the traits of the elements were governed by polygenic controls (Supplementary Fig. S3).

Distinct Correlation Patterns of Elemental Concentrations in Grain and Straw

To investigate the relationship between the elemental accumulation in straw and grain, we performed Pearson’s correlation analysis of the concentrations of the 13 elements in the MAGIC population (Supplementary Table S6). The results reveal a wide range of correlation coefficients (r), indicating varying degrees of element translocation and partitioning between the vegetative and reproductive organs. Among these, Ca, As, and Mo exhibited relatively strong positive correlations. Other elements exhibited weak or negligible correlations.

Next, we calculated the correlation coefficients for all pairs of the 13 elements in grain and straw. In grain (Fig. 1A), days to heading were positively correlated with Cd (r = 0.68), Ni (r = 0.46), and Mn (r = 0.39), and negatively correlated with As (r = −0.59), Mo (r = −0.45), Fe (r = −0.37), Zn (r = −0.30), Mg (r = −0.32), and

Fig. 1.

Correlation coefficient matrix between the 13 elements and days to heading in the MAGIC population. Matrix heat maps were made based on the data for (A) grain and (B) straw. According to the Pearson coefficient (r), red indicates a positive correlation, while green indicates a negative correlation. The intensity of the color reflects the strength of the correlation. The bottom-left and top-right sections of the matrix display the correlation coefficients and significance levels (****: 0.0001, ***: 0.001, **: 0.01, and *: 0.05, respectively).

P (r = −0.32). Among the elemental pairs, the strongest positive correlation was found between Mg and P (r = 0.86), followed by Zn–P (r = 0.70), Zn–Fe (r = 0.65), and Cd–Ni (r = 0.62). In addition, significant positive correlations were observed between Cu and Zn (r = 0.56), Cu and Fe (r = 0.54), and Mn and Ca (r = 0.58). Conversely, distinct negative correlations were evident between Cd and As (r = −0.55), Cd and Mo (r = −0.34), and As and Ni (r = −0.44). These correlation trends were consistent with the grain data from 2019 (Supplementary Fig. S4).

In the straw (Fig. 1B), the correlation patterns differed partially from those observed in the grain, although some similarities existed. For instance, correlations between days to heading and both Cd (r = −0.02) and Ni (r = −0.02) were negligible in straw, whereas negative correlations with Mo (r = −0.46) and As (r = −0.37) were like those observed in grain. Generally, positive correlations were dominant among the element pairs, with the highest observed between Cr and Ni (r = 1.00), followed by Cr–Cu (r = 0.74) and Cr–Fe (r = 0.71). A clear correlation cluster was identified between Fe, Ni, and Cu (Fe–Ni: r = 0.70, Ni–Cu: r = 0.73, and Fe-Cu: r = 0.58). Mn exhibited a distinct positive correlation with Cd (r = 0.62) and Mg (r = 0.56). Importantly, the correlation between Mg and P (r = 0.36) was markedly weaker compared with grain (r = 0.86). Moreover, strong negative correlations found in grain, such as Cd–As, were only weakly negative (r = −0.11) in straw, further highlighting sample-type-specific relationships.

The integration of these findings indicates that the correlation between “elements” (or between “elements” and “days to heading”) manifests differently across sample types. It is noteworthy that elements such as Cd, Mg, and P demonstrate divergent correlations with other elements when comparing straw and grain.

QTLs Detection in the MAGIC Population

At a GWAS significance threshold of FDR ≤ 0.1, 53 QTLs for five elements (Cd, Cu, Fe, Mn, Mo) in grain, and 51 QTLs for four elements (Ca, Mg, Mn, Mo) in straw were detected (Fig. 2; Table 1). Specifically, grain QTLs included Cd (15 loci), Cu (9), Fe (19), Mn (1), and Mo (9), whereas straw QTLs included Ca (11), Mg (20), Mn (9), and Mo (11). Although QTLs for Mn and Mo were found in both the grain and the straw, no common regions were found. No QTLs were detected for other elements in the straw (P, K, As, Cd, Cr, Cu, Fe, Ni, and Zn) or grain (P, K, Ca, Mg, As, Cr, Ni, and Zn) at the specified thresholds (Supplementary Fig. S5). The detected QTLs were distributed across all the chromosomes, with the highest density observed on chromosome 6, where 19 QTLs were detected for seven elements (Ca, Cd, Cu, Fe, Mg, Mn, and Mo). In contrast, only four QTLs for the three elements were found on chromosomes 8 (Mo, Mg, and Fe), 9 (Cd, Mg, and Mo), and 10 (Mg, Ca, and Mo) (Fig. 3; Table 1). Nineteen QTL clusters formed by overlapping multiple significant QTL regions were observed across all chromosomes, except for chromosomes 5 and 12 (Fig. 3; Supplementary Table S7).

Fig. 2.

Manhattan plots of the significant QTLs. The left and right panels show Manhattan plots for grain and straw, respectively. The horizontal axis indicates the chromosome names, and the vertical axis shows the log₁₀(P) value. The red and blue horizontal lines denote the significance thresholds at FDR = 0.1 and FDR = 0.05, respectively. Data points exceeding the FDR threshold of 0.1 are highlighted in red.

Fig. 3.

Distributions of the 104 significant QTLs on rice physical map. The physical positions were referenced from IRGSP 1.0 (Kawahara et al. 2013). A threshold of FDR ≤ 0.1 was applied. The numbers on the left side of each chromosome indicate the physical positions (Mb) of 342 representative SNP loci selected at approximately equal distances from the 13,603 SNP loci. The bars on the right side of each chromosome represent the QTL regions, with different colors denoting various elements. In the QTL names, the prefix “q” is followed by “21” to indicate the sampling year (2021), “G” for grain, and “S” for straw. The grey shaded region indicates a QTL cluster formed by the partial overlap of multiple QTLs, and the gray letters beginning with “C” indicate the corresponding QTL cluster in Supplementary Table S7.

To identify more significant regions from a total of the 104 detected QTLs, the threshold was tightened to FDR ≤ 0.05, reducing the total number of significant QTLs to 10 across grain and straw (Table 1). In the grain, six Cd QTLs were detected on chromosomes 3, 6, 7, and 11, and four Mo QTLs were identified on chromosomes 7, 8, and 12. In straw, five Ca QTLs were found on chromosomes 2, 6, and 11, one Mn QTL on chromosome 5, and four Mo QTLs on chromosomes 1, 6, 10, and 11. The most significant QTLs included q21SMo1 (-log₁₀(p) = 6.39), q21GMo7-1 (5.36), q21GCd6-2 (4.58), and q21SMn5 (4.52) (Table 1).

Prioritization of Candidate Genes Based on Haplotype Information

Among the genes screened using the keywords, we compared founder haplotypes and phenotypic data for 430 genes located within the detected 104 QTL regions. In total, 52 genes distributed across 15 QTLs related to the six elements exhibited concordant changes in both haplotypes and phenotypes (Table 2).

Table 2.

Candidate genes estimated from the consistency between phenotypic values and haplotype variations of the eight founder varieties

Element	QTL name	Chr.	PVE*1)	Gene	OsID	Description and (Ref#)*2)	Hap1	Hap2	Hap3	Most probable SnpEff	Nucleotide and amino acid change (from AK Hap to others)	Sup. Fig.S6*3)
	q21SCa2-1	2	11.2	OsACA9	Os02g0176700	Similar to autoinhibited calcium ATPase (Meng et al. 2015)	HO, RU, SU	TK	AK, BE, MI, TC	missense_variant	c.58G > A|p.Asp20Asn	Figure 4. DEF
				OsHSL3	Os02g0320800	Similar to Iron/ascorbate-dependent oxidoreductase (Kawahara et al. 2013)	HO, RU, SU, TK	AK, BE, MI, TC		stop_lost	c.907T > C|p.Ter303Argext*?	1
Ca	q21SCa2-3	2	13.7	OsCAX1c	Os02g0314300	Sodium/calcium exchanger membrane region domain containing protein (Huynh et al. 2018)	HO, RU, SU, TK	AK, BE, MI, TC		missense_variant	c.409 A > C|p.Ile137Leu	2
				OsPP2C14	Os02g0471500	Protein phosphatase 2 C domain containing protein (Kover et al. 2009)	HO, RU, SU	TK	AK, BE, MI, TC	missense_variant	c.605G > A|p.Ser202Asn	3
	q21SCa11	11	9.0	OsCML6	Os11g0586200	Similar to Calmodulin (Bandillo et al. 2013)	HO, RU, SU, TK	BE, TC	AK, MI	splice_donor_variant&intron_variant	c.326 + 1G > A	Figure 4. GHI
	q21GCd6-2	6	15.8	OsWAK61	Os06g0142500	Similar to Wall-associated kinase 3 (Pasion et al. 2023)	RU	AK, BE, HO, MI, SU, TC, TK		frameshift_variant&start_lost	c.−9_17delGGCCATCCGATGGGAGCCTTAGTACT	4
				OsCDT2	Os06g0143100	Similar to Cadmium tolerant 1 (Huang et al. 2012)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.27 C > A|p.Asp9Glu	5
				SPDT	Os06g0143700	SULTR-like phosphorus distribution transporter (Ogawa et al. 2018)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.140 C > T|p.Ala47Val	6
				OsPLL6	Os06g0144200	Similar to Pectate lyase homolog (EC 4.2.2.2) (R Core Team 2018)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.1321 A > G|p.Thr441Ala	7
Cd				OsPLL7	Os06g0144900	Pectate lyase/Amb allergen domain containing protein (R Core Team 2018)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.976 C > A|p.Arg326Ser	8
	q21GCd7-1	7	9.9	OsZIP8	Os07g0232800	Zinc transporter, Zn uptake and distribution (Huang et al. 2019)	HO, MI, RU, SU	TK	AK, BE, TC	conservative_inframe_deletion	c.481_486delACCGCC|p.Thr161_Ala162del	9
				HMA3	Os07g0232900	Root-to-shoot cadmium (Cd) translocation (Ma and Yamaji 2006)	HO, RU, TK	MI, SU	AK, BE, TC	missense_variant		10
				OsNramp1	Os07g0258400	Integral membrane protein, Metal ion transport (Cavanagh et al. 2008)	HO, MI, RU, SU, TK	AK, BE, TC		splice_acceptor_variant&3_prime_UTR_variant&intron_variant	c.*298-1_*339delAAATATAAATGTAAATAAATGCAGTTAAAGTAACACCTTGCTC	11
Cu	q21GCu2	2	13.9	OsPP2C22	Os02g0607500	Putative protein phosphatase 2 C 22 (Ogawa et al. 2018)	RU	HO, TK	AK, BE, MI, SU, TC	stop_gained	c.1348G > T|p.Glu450*	12
	q21GCu7	7	9.0	OsTFIIS	Os07g0229700	Similar to DNA-directed RNA polymerase I subunit 12 (Kathpalia 2018)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.250G > A|p.Ala84Thr	13
				OsCYB5-14	Os07g0232200	Similar to Flavohemoprotein b5/b5R variant (Li et al. 2023)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.157G > A|p.Ala53Thr	14
				DRM2	Os03g0110800	DNA methyltransferase, Vegetative and reproductive development (Khush 2005)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.1217 A > G|p.His406Arg	15
				OsF5HL2	Os03g0112900	Similar to Aldehyde 5-hydroxylase (Hawkesford 2012)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.427G > C|p.Val143Leu	16
				TK	Os03g0113100	Similar to Thymidine kinase (Mackay et al. 2014)	RU	AK, BE, HO, MI, SU, TC, TK		disruptive_inframe_insertion	c.161_163dupCGG|p.Ala54dup	17
	q21GFe3	3	12.0	OsCd1	Os03g0114800	Major facilitator superfamily protein, Cd uptake, Grain Cd accumulation (Orasen et al. 2025)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.1346T > A|p.Val449Asp	18
				OsENODL10	Os03g0115000	Cupredoxin domain containing protein (Dell'Acqua et al. 2015)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.344T > C|p.Leu115Pro	19
Fe				OsHIPP27	Os03g0126700	Similar to Barley stem rust resistance protein (Codex 2025)	RU	AK, BE, HO, MI, SU, TC, TK		disruptive_inframe_insertion	c.346_347insACGACGGCGACGGCAAGAAGG|p.Lys115_Gly116insAspAspGlyAspGlyLysLys	20
				CIPK9	Os03g0126800	Regulation of ammonium-dependent root growth (Miyadate et al. 2011)	RU	AK, BE, HO, MI, SU, TC, TK		frameshift_variant	c.1243delC|p.Leu415fs	21
				OsCDPK7	Os03g0128700	Calcium-dependent protein kinase, isoform 11 (EC 2.7.1.-) (CDPK 11) (Pascual et al. 2015)	RU	AK, BE, HO, MI, SU, TC, TK		missense_variant	c.1420G > A|p.Val474Ile	22
				OsCYB5-14	Os07g0232200	Similar to Flavohemoprotein b5/b5R variant (Li et al. 2023)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.157G > A|p.Ala53Thr	23
	q21GFe7-2	7	11.3		Os07g0291750	Electron carrier/heme binding/iron ion binding/monooxygenase	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.113G > A|p.Arg38Lys	24
					Os07g0292100	Similar to CYP71B6 (CYTOCHROME P450 71B6)	HO, MI, RU, SU, TK	AK, BE, TC		frameshift_variant	c.388_394delAAACTGG|p.Lys130fs	25
Mg	q21SMg3-3	3	13.2	2-ODD26	Os03g0856000	2OG-Fe(II) oxygenase domain containing protein (Boonburapong and Buaboocha 2007)	HO, MI, RU, SU, TK	AK, BE, (TC)		frameshift_variant	c.306_307delCG|p.Asp103fs	26
				OsHMP23	Os03g0861400	Hypothetical conserved gene (Codex 2025)	HO, MI, RU, SU, TK	AK, BE, (TC)		disruptive_inframe_insertion	c.990_992dupCCA|p.His330dup	27
	q21SMg8-1	8	10.9	OsP0A	Os08g0130500	Similar to 60 S acidic ribosomal protein p0 (Chang et al. 2020)	HO, RU, TK	AK, BE, TC, MI, SU		missense_variant	c.770 A > G|p.Glu257Gly	28
				OsTFIIS	Os07g0229700	Similar to DNA-directed RNA polymerase I subunit 12 (EC 2.7.7.6) (Kathpalia 2018)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.250G > A|p.Ala84Thr	29
				OsCYB5-14	Os07g0232200	Similar to Flavohemoprotein b5/b5R variant (Li et al. 2023)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.157G > A|p.Ala53Thr	30
				HMA3	Os07g0232900	Root-to-shoot cadmium (Cd) translocation (Ma and Yamaji 2006)	HO, RU, TK	MI, SU	AK, BE, TC	missense_variant		31
				OsNramp1	Os07g0258400	Similar to Metal transporter (Cavanagh et al. 2008)	HO, MI, RU, SU, TK	AK, BE, TC		splice_acceptor_variant&3_prime_UTR_variant&intron_variant	c.*298-1_*339delAAATATAAATGTAAATAAATGCAGTTAAAGTAACACCTTGCTC	32
	q21GMo7-1	7	12.0	OsC3H49	Os07g0281000	Nucleotide-binding, alpha-beta plait domain containing protein	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.1097 C > A|p.Thr366Lys	33
				OsPARP1	Os07g0413700	Response to DNA damage, gametophyte development, promotion of seed setting rate (Islam et al. 2016)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.138 A > C|p.Leu46Phe	34
				OsCesA3	Os07g0424400	Similar to Cellulose synthase-7 (Mangino et al. 2022)	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.537G > A|p.Met179Ile	35
				XBOS34	Os07g0446100	Zinc finger, RING/FYVE/PHD-type domain containing protein	HO, MI, RU, SU, TK	AK, BE, TC		missense_variant	c.887T > A|p.Leu296Gln	36
Mo	q21GMo8	8	13.1	OsMOT1;1	Os08g0101500	Molybdate transporter, Uptake and translocation of molybdate (Gebert et al. 2009)	HO	SU(Hap2),TK(Hap3),RU(Hap4),TC(Hap5)	AK, BE, MI(Hap6)	frameshift_variant	c.642_643delGC|p.Gln215fs c.646_652delCAGCAGC|p.Gln216fs	Figure 4. ABC
				NADP-ME2	Os01g0723400	NADP-malic enzyme, salt and osmotic stress tolerance, disease resistance (Kumagai et al. 2019)	HO, RU, SU, TK	BE	AK, MI, TC	frameshift_variant	c.19_20insCG|p.Val7fs	37
				RD22	Os01g0733500	Similar to Dehydration-induced protein RD22-like protein 1 (Descalsota-Empleo et al. 2019)	RU, HO, TK, SU	AK, BE, MI, TC		missense_variant	c.1009G > A|p.Ala337Thr	38
				OsPFK02	Os01g0738800	Conserved hypothetical protein (Kojima et al. 2002)	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.77T > C|p.Val26Ala	39
					Os01g0739000	Similar to Mitochondrial processing peptidase	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.50 A > C|p.Tyr17Ser	40
				OscytME3	Os01g0743500	Cytosolic NADP malic enzyme (Chen et al. 2022)	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.796 A > C|p.Asn266His	41
	q21SMo1	1	15.4	OsGATA6	Os01g0745700	Positive regulation of panicle development and grain number, regulation of grain size (Pingali 2012)	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.994 C > T|p.His332Tyr	42
				NUP85	Os01g0746200	Nucleoporin, Common symbiosis signaling (SYM) pathway (Duan et al. 2017)	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.1243G > A|p.Gly415Arg	43
				OsCESA4	Os01g0750300	Cellulose synthase catalytic subunit, secondary cell wall formation (International Rice Genome Sequencing Project 2005)	HO, RU, TK, SU	AK, BE, MI, TC		missense_variant	c.221 C > T|p.Ala74Val	44
				NBIP1	Os01g0755700	RING-type E3 ubiquitin ligase, Regulation of phosphate and nitrate signalling (Evenson and Gollin 2003)	RU, TK	AK, BE, HO, MI, SU, TC		missense_variant	c.233T > C|p.Val78Ala	45
				OsHMP6	Os01g0758000	Similar to copper-binding family protein (Codex 2025)	RU	HO, TK, SU	AK, BE, MI, TC	disruptive_inframe_deletion	c.297_305delACACGCAGC|p.His100_Ala102del	46
				OsCIPK30	Os01g0759200	CBL-interaction protein kinase protein, Resistance to Rice stripe virus (RSV) disease (Ji et al. 2022)	RU	HO, TK, SU	AK, BE, MI, TC	missense_variant	c.499G > A|p.Gly167Ser	47
				OsPK7	Os01g0759400	OsPK7 (Os01t0759400-02)	RU, HO, TK, SU	AK, BE, MI, TC		missense_variant	c.780T > G|p.Asn260Lys	48
	q21SMo10	10	13.6	OsHIPP25	Os10g0532300	Heavy metal transport/detoxification protein domain containing protein (Codex 2025)	SU	AK, BE, HO, MI, TC, TK, RU		disruptive_inframe_deletion	c.270_293delCGGCGGCGGCGACGGCAAGAAGGA|p.Asp90_Lys97del	49

^*1)PVE (phenotypic variance explained) was calculated as the coefficient of determination (R²) from a linear model fitting the phenotype to the marker haplotype/genotype (phenotype ~ marker). PVE is reported as 100 × R²

^*2)The reference papers supporting the description are listed under their corresponding numbers in Supplementary Information S8

^*3)The figure index of results of multi-haplotype comparison in Supplemental Fig. S6

Several of the 52 candidate genes have been reported to be associated with elemental accumulation. Among them, the q21GMo8 region includes OsMOT1;1 (Os08g0101500), which encodes a molybdate transporter significantly associated with Mo accumulation (Huang et al. 2019). We first selected the SNP site closest to OsMOT1;1 and performed multiple comparison analysis using the haplotype of the eight founders and the Mo concentration data of 182 lines in the MAGIC population (Fig. 4A). Subsequently, sequence variations in OsMOT1;1 in the eight founders were analyzed using TASUKE+. Two deletions (2 bp at positions 642–643 and 7 bp at positions 646–652) were identified in the transcript of OsMOT1;1 specifically in the founder HO, both annotated as high-impact frameshift mutations by SnpEff. (Fig. 4B). In addition, moderate-impact mutations were detected in HO, RU, SU, TC, and TK, whereas no moderate- or higher-impact mutations were found in AK, BE, or MI. Based on these variation patterns, we grouped the eight founders into six haplotypes, Hap1 (HO), Hap2 (SU), Hap3 (TK), Hap4 (RU), Hap5 (TC), and Hap6 (AK, BE, and MI, respectively, with no more than moderate mutations), followed by a multiple comparison analysis (Fig. 4C). The results show that Hap1 had a significantly lower Mo content than Hap6, whereas the Mo content of Hap2 to Hap5 (carrying moderate-impact mutations) showed intermediate values. These data suggest that natural variations in OsMOT1;1 affect the accumulation of Mo in grain.

Fig. 4.

Genetic structure and multiple haplotype comparison for OsMOT1;1, OsACA9, and OsCML6 that are candidate genes of the three QTLs: q21GMo8, q21SCa2-1, and q21SCa11. (A), (D), and (G) show the results of multiple comparisons among eight haplotypes. The red rhombus in each boxplot represents the mean value of the group. The red letters indicate statistically significant differences. (B), (E), and (H) show the gene structure and haplotype variations of OsMOT1;1, OsACA9, and OsCML6, respectively. Yellow and red arrows represent SNPs and the corresponding diagonal squares indicate deletions with moderate and high predicted SnpEff effects, respectively. (C), (F), and (I) show the results of multiple comparisons among haplotype groups based on sequence variation. The meanings of the red rhombus and letters are the same as those in (A), (D), and (G).

Within the q21SCa2-1 region, we predicted OsACA9 (Os02g0176700), a novel candidate gene encoding an auto-inhibited calcium ATPase that may be involved in Ca transport (Wang X et al. 2024). Using the same approach, we performed founder haplotype-based analysis (Fig. 4D). In the investigation of sequence variation across the founders, six moderate-impact variants were identified in HO, RU, and SU (Hap1); three moderate SNPs were detected in TK (Hap2); and no moderate or higher impact mutations were found in AK, BE, MI, and TC (Hap3) (Fig. 4E). Multiple comparisons across the three haplotypes revealed that Hap1 had a significantly lower Ca content than Hap3, whereas that of Hap2 was intermediate (Fig. 4F). This result supports OsACA9 potentially involved in the natural variation in Ca accumulation in straw.

Interestingly, the candidate genes listed in Table 2 also include genes not previously associated with the relevant elements. For example, within q21SCa11, we predicted OsCML6 (Os11g0586200), a gene encoding a calmodulin-like protein (Boonburapong and Buaboocha 2007), was a potential regulator of Ca accumulation. Multiple comparison analyses showed that AK and MI had significantly lower Ca concentrations than the other treatments (Fig. 4G). High-impact splice donor variants were detected in BE, HO, RU, SU, TC, and TK (Fig. 4H). Four additional moderate-impact SNPs were also identified. Further analysis revealed that a moderate SNP at position 348 bp was present in HO, RU, SU, and TK, whereas an SNP at 47 bp was present in BE and TC. Therefore, we grouped the founders as Hap1 (HO, RU, SU, and TK), Hap2 (BE and TC), and Hap3 (AK and MI without moderate- or high-impact variants) (Fig. 4H). The comparison among haplotype groups revealed that Hap1 and Hap2, which carry splice donor or moderate-impact mutations, exhibited significantly higher Ca content than Hap3, with no significant difference between Hap1 and Hap2 (Fig. 4I). This result implies that the splice donor variant of OsCML6 may enhance gene function or expression by altering splicing efficiency, thereby increasing Ca accumulation in the straw.

Taken together, the integration of SnpEff annotation, sequence variant analysis, and haplotype–phenotype association of multiple founders enables the MAGIC population to accurately narrow down candidate genes associated with elemental accumulation. The results of all the other candidate genes are shown in Supplementary Figs. S6-1 to S6-49.

Discussion

Effectiveness and Considerations for the MAGIC Population

Multi-parent advanced generation inter-cross (MAGIC) populations provide an effective platform for dissecting complex quantitative traits because they combine multiple founder genomes and accumulate recombination events across generations. Compared with conventional biparental populations and diversity panels, MAGIC populations increase genetic diversity caused by recombination among multiple founder alleles. This improves the resolution of genetic dissection of complex traits. Consistent with previous studies using MAGIC populations (Wang P. et al. 2024; Zhi et al. 2023), our MAGIC population redetected known genetic loci and highlighted additional candidate regions. For example, q21GMo8 co-localized with the previously reported Mo transporter gene OsMOT1;1, and q21SCa2-1 mapped near the Ca-related ATPase OsACA9. Importantly, we also detected loci that have not been emphasized in prior studies, such as q21SCa11, in which OsCML6 represents a plausible candidate gene (Fig. 4).

It is important to note that because eight founder alleles were assumed in this MAGIC population, the number of lines representing each genotype was smaller than that in a biparental population of the same size. Consequently, the statistical power to detect QTLs may be reduced, resulting in higher p-values despite similar levels of phenotypic variation. To reduce the risk of false negatives, we applied a relatively permissive significance threshold (FDR ≤ 0.10) during the initial QTL detection stage (Table 1). Candidate genes were subsequently prioritized using haplotype comparisons and polymorphism analyses among founder alleles. Although this does not negate the ability of the MAGIC population to perform GWAS at a high resolution, larger population sizes are preferable for sensitivity and detection purposes.

Relationship Between Element Accumulation and Days to Heading Based on QTL Distribution

The accumulation of elements in rice is influenced by both the genetic background and environmental factors, among which flowering time (i.e., days to heading) is a key trait affecting elemental accumulation (Duan et al. 2017; Descalsota et al. 2019). In the present study, days to heading were significantly positively correlated with grain Cd concentration (r = 0.68, Fig. 1A). Previous findings have suggested that late-flowering rice genotypes tend to accumulate more Cd (Yamamoto et al. 2024), likely because of extended root uptake periods and prolonged element translocation.

Further, QTL mapping supported a genetic contribution to the observed association (Figs. 2 and 3). At the short-arm end of chromosome 6, two QTLs for grain Cd accumulation (q21GCd6-1 and q21GCd6-2) were mapped close to the flowering-time regulators Hd3a (Os06g0157700) and RFT1 (Os06g0157500) (Kojima et al. 2002). However, grain Cd accumulation is strongly influenced by the bioavailability of Cd in soil and associated field conditions. This proximity does not necessarily indicate a direct causal role of flowering genes in Cd transport or accumulation. Instead, the signal may reflect linkage with nearby Cd-related loci, or an indirect effect whereby heading date alters the timing and duration of plant exposure to soil Cd availability and redox dynamics during grain filling. In addition, several other elements showed a QTL cluster in the same region on chromosome 6 (Fig. 3; Supplementary Table S7), including Mo (q21SMo6-1, q21SMo6-2), Fe (q21GFe6-1), Cu (q21GCu6-1), Mn (q21SMn6-1), Mg (q21SMg6-1), and Ca (q21SCa6-1). Collectively, these patterns are consistent with a genomic region enriched for element QTLs, although whether this reflects shared regulators, tightly linked loci, or pleiotropic effects mediated by developmental or environmental responsiveness requires further validation.

A similar trend was observed for chromosome 7. The grain Cd QTL (q21GCd7-1) mapped near Ghd7 (Os07g0261200), a major regulator of days to heading and plant architecture (Xue et al. 2008). As discussed above, this proximity could reflect either a tight linkage to Cd-related loci or an indirect association mediated by developmental timing. In addition, QTLs for Mo (q21GMo7-1), Fe (q21GFe7-2), Cu (q21GCu7), and Mg (q21SMg7-1) were identified in this region (Fig. 3; Supplementary Table S7). This suggests the existence of a genetic cluster that jointly influences the days to heading and the accumulation of multiple elements.

Relationship of Element Accumulation Between Grain and Straw Based on QTL Distribution

We demonstrated a clear difference in the correlation structures between straw and grain (Fig. 1), mapped 104 QTLs to the rice genome (Fig. 3), and grouped them into 19 QTL clusters (Supplementary Table S7). Integrating these data revealed that several element pairs showing high correlations within the sample significantly overlapped with specific QTL clusters. These results suggest that these regions harbor genetic factors that jointly regulate transport, allocation, and specific accumulation.

In the straw, the positive correlation of Mo–Ca (chromosome 1, C1), moderate correlation of Mn–Ca (chromosome 2, C2-1), and strong correlation of Mn–Mg (chromosomes 6, C6) coincided with their corresponding QTL clusters. This phenotypic correlation and genetic co-localization pattern suggest that these regions may contain pleiotropic genes that simultaneously regulate the accumulation of two elements or tightly linked loci with coordinated effects. Notably, the C6-5 region on chromosome 6 is a QTL-enriched hotspot and overlaps with a strong Mn–Mg correlation region. This region contains candidate genes related to Mg transport, such as members of the MRS2/MGT family (Gebert et al. 2009), suggesting a potential mechanistic link. Nevertheless, direct evidence for a shared transporter or causal gene affecting both elements is still lacking and requires further validation.

In contrast, a different pattern was observed in the grain. The C7 region on chromosome 7 showed positive correlations of Mo–Fe and Mo–Cu, along with a negative correlation of Mo–Cd. This complex relationship is likely attributable to the combined actions of developmental regulators and metal transporters. As mentioned in the previous section, the Ghd7 gene, which is involved in the days to heading, may have contributed to this. In addition, this region contains major metal transporters. OsHMA3 (Os07g0232900), a vacuolar Cd transporter, determines Cd sequestration, and loss-of-function alleles significantly increase Cd translocation to grain, whereas functional alleles can effectively reduce Cd accumulation (Miyadate et al. 2011; Ueno et al. 2010). Furthermore, OsNRAMP5 (Os07g0257200), a major transporter of Mn/Cd uptake in the roots, is induced under Fe deficiency (Chang et al. 2020; Sasaki et al. 2012). This provides a mechanistic explanation for the antagonistic relationships among Mo, Fe, and Cd. Thus, the C7 cluster can be considered a regulatory hub. Here, Ghd7-mediated developmental effects determine the source–sink balance, whereas specific metal transporters govern the selective partitioning of ions. This mechanistic separation likely underlies the observed Mo–Fe–Cu synergy and Mo–Cd antagonism.

Validity of the Estimated Candidate Genes

This study integrated SnpEff annotation, structural variant analysis, and haplotype–phenotype association to predict 52 candidate genes for 15 QTL regions associated with six elements. The MAGIC population, which includes multiple founder SnpEff information, was well-suited for this integrated analysis, enabling the prioritization of known and novel candidate genes underlying elemental accumulation. These candidates were related to six elements: Ca (5 genes), Cd (8), Cu (3), Fe (11), Mg (3), and Mo (22) (Fig. 4; Table 2; Supplementary Fig. S6). This strategy not only confirmed several previously reported functional genes, such as OsMOT1;1 and OsACA9, but also identified novel candidates, including OsCML6, thereby offering new insights into the genetic regulation of elemental accumulation in rice. For OsMOT1;1 and OsACA9, distinct types of mutations—including frameshift and missense variants—were significantly associated with Mo and Ca concentrations, respectively (Fig. 4ABC; Fig. 4DEF). Notably, the splice donor site variant of OsCML6 was positively correlated with Ca levels, suggesting its possible role in modulating gene expression through altered splicing efficiency (Fig. 4GHI), thereby influencing Ca homeostasis.

Two well-characterized heavy-metal transporter genes, OsHMA3 (Supplementary Fig. S6–10), and OsNRAMP1 (Os07g0258400) (Supplementary Fig. S6–11) were estimated within the Cd-associated QTL region on chromosome 7 (q21GCd7-1). Haplotype analysis of the eight MAGIC parents revealed structural variants of OsHMA3 in HO, MI, RU, SU, and TK, whereas AK, BE, and TC lacked moderate- or high-impact mutations. Multiple comparisons showed that lines carrying the mutated alleles exhibited significantly higher grain Cd concentrations, supporting previous findings that loss-of-function mutations in OsHMA3 enhance Cd translocation to the grain. A similar trend was observed for OsNRAMP1. Collectively, this study not only validated known metal transporter genes under natural allelic variation but also proposed novel regulatory candidate alleles.

Application of the Rice MAGIC Population for Improving Elemental Accumulation Under Different Soil Conditions

Our MAGIC population, consisting of multiple founder alleles, is particularly well-suited for broadly detecting genotype × environment (G×E) interactions within cultivated rice species. Although the present study was conducted under single soil conditions, phenotypic assessments and GWAS of MAGIC across diverse soil environments are expected to advance our understanding of the mechanisms underlying elemental accumulation in nature. For example, Orasen et al. (2025) conducted an ionomics GWAS under contrasting water management conditions, revealing that most marker–trait associations were specific to the conditions. We also evaluated the accumulation patterns of 13 elements in straw at the heading stage and in grain at maturity under acidic, neutral, and alkaline soil pH conditions using eight founders of this MAGIC population (Yamamoto et al. 2024). This experiment suggests that there are element-specific interactions between genotype, soil pH, and flowering time, highlighting the importance of investigating the genetic basis underlying these interactions. Building upon these findings, performing a GWAS of MAGIC across contrasting soil conditions could provide a more robust genetic foundation for understanding nutrient adaptation under stressful environments in rice and support the identification of genes with broad-spectrum or environment-specific advantages.

Furthermore, from the perspective of gene network research on element accumulation, utilizing MAGIC populations with high haplotype resolution holds the potential to rapidly translate large-scale ionomics information into practical breeding objectives (Tiozon et al. 2024). Furthermore, integrating predicted candidate gene sets into regulatory networks or gene-element network models may reveal shared pathways between elements and potential hub regulators (Pasion et al. 2023). Overall, this study provides a dataset and analytical framework that serve as a foundation for molecular breeding.

Conclusions

This study highlights the unique advantages of multi-parental MAGIC populations for dissecting complex traits such as elemental accumulation in rice. Unlike biparental populations, MAGIC offers broad genetic diversity and high recombination, enabling the detection of moderate-effect loci and improving mapping resolution. We provided a comprehensive overview of genetic variation in the uptake and accumulation of 13 elements in rice grain and straw, revealing extensive phenotypic diversity and sample-type-specific correlations. By integrating GWAS with haplotype-based analysis, we predicted candidate genes involved in the uptake and partitioning of essential and toxic elements, including both known transporters and novel regulators. These findings provide a strong genetic foundation for breeding programs aimed at enhancing nutrient efficiency and stress tolerance. Future work should focus on functional validation of these genes and their incorporation into genomic selection strategies to accelerate rice improvement under diverse environmental conditions.

Abbreviations

AK

Akidawara

As

Arsenic

B

Boron

BE

Bekogonomi

C

Carbon

Ca

Calcium

Cd

Cadmium

Cl

Chlorine

Cr

Chromium

Cu

Copper

CV

Coefficient of variation

FDR

False discovery rate

Fe

Iron

GWAS

Genome-wide association study

G×E

Genotype by environment interaction

Hg

Mercury

HO

Hokuriku 193

ICP-MS

Inductively coupled plasma mass spectrometry

K

Potassium

MAGIC

Multi-parent advanced generation inter-cross

Mg

Magnesium

MI

Mizuhochikara

Mn

Manganese

Mo

Molybdenum

N

Nitrogen

Ni

Nickel

OsID

Oryza sativa ID

P

Phosphorus

Pb

Lead

QTL

Quantitative trait loci

RU

Ruriaoba

Si

Silicon

SNP

Single nucleotide polymorphism

SU

Suweon 258

TC

Tachiaoba

TK

Takanari

Zn

Zinc

Author Contributions

TY designed the study. QZ, TY, and KK conducted the experiments. DO, JY, and TY developed the plant materials. QZ, TF, DO, JY, and TY analyzed the data. JFM assisted and supervised ICP-MS experiments. QZ wrote the manuscript and TY edited and revised it. All the authors have read and approved the final version of the manuscript.

Funding

This work was supported in part by JSPS KAKENHI Grants-in-Aid for Scientific Research 20H02958 to TY, 16H06296 and 21H05034 to JFM, the China Scholarship Council Program (Project ID: 202305080109), and the Ohara Scholarship Foundation.

Data Availability

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Ethics Approval and Consent to Participate

Not applicable.

Consent for Publication

Not applicable.

Competing Interests

The authors declare no competing interests.