The control of macro- and micro-nutrient homeostasis in plants have been extensively studied, however the loci that control natural ionomic variations in the grain are still largely undetermined in rice 31,32,33. GWAS has been a powerful to dissect complex traits in plants31, however there are several factors that can limit the success of using GWAS to study the rice ionomes. These limiting factors include the relatively little variation in plant elemental concentrations and the often-substantial environmental effect34,35. In plant the the transport and homeostasis of essential mineral nutrients are highly regulated processes as they require adequate levels of these essential nutrients for their growth and reproduction, while at the same time excess accumulation can also be detrimental to cell growth36.
The rice panel used in this study represents an excellent resource for genetic diversity covering a wide geographical and ecological variation in rice germplasm25,26,30. This diversity promising a large number of haplotypes is advantageous, however the effects of population structure need to be accounted for, in this case through a mixed model approach. The high density of SNPs (∼17 SNPs per kb on average) in our GWAS panel also facilitates high-resolution mapping with the loci were generally obtained within ∼100 kb, much higher than those obtained using linkage mapping approach37. This high resolution makes it possible to identify plausible candidate genes for a number of loci identified by GWAS using the information about the functional gene annotation or their orthologous genes in other plant species38,39.
The group of elements that showed relatively low levels of variation for the grain element concentrations consisted of four essential macronutrients (K, Mg, P, Ca) and four essential micronutrients (Cu, Fe, Mn and Zn) (Table 1, FigS1). These results indicate that the homeostasis of these elements is under relatively tight regulation. Previous research has shown that plants have evolved regulatory mechanisms to control the internal fluctuation of the essential nutrients to maintain their concentrations within narrow ranges for optimal growth, development and seed production40,41. Significantly larger variations were found for the concentration of the second group consisting of the three essential micronutrients B, Co and Mo and Na (Na is a functional but nonessential element)42. It is likely that the elements in the second group were under less pressure to regulate their concentrations (unless they approach toxicity levels), thus having more relaxed control mechanisms. The differences in these control mechanisms exist not only among genotypes, they can also vary temporally and spatially within a given plant. Because this regulatory variability exists, it would appear that enhancing the micronutrient density of edible plant components through the manipulation of physiological processes is an achievable goal. The high heritability values of the nine grain elements also indicate that the contribution of genotypic variance to the total phenotypic variance was significant for these traits. Similar results were reported in previous studies11,43,44. Thus, direct selection for these elements may be a practical approach for trait improvement.
All twelve elements concentrations in the grain were negatively correlated with grain yield in at least two environments (Supp. Table S2). Six elements including Fe, K, Mg, Na, P and Zn had negative correlations with grain yield in all four environments and the highest correlation coefficients were found with Fe, Mg and P (r: -0.39 to -0.52). The negative correlations between grain yield and grain element concentrations are not uncommon in rice and have been reported in past studies for K, Mg, Mn, P, and Zn11,20. This likely reflects the dilution effects of increasing grain mass on the elemental concentrations. Minimizing the effect of grain yield for genetic mapping may be a required corrective measure in determining the genetics controlling this element accumulation in the grain, which would benefit breeding for rice lines with high nutrient concentrations. In our study, despite having strong negative correlations with all elements, grain yield had only one co-located QTL with Mo concentration on chr 10 (16.8Mb) in one environment (PR13). This suggests that selection to enhance these grain elements at the identified loci is not likely to incur a yield penalty.
PH had consistent positive correlations with Co, Ca K and Zn; negative correlations with Cu and Mo and no correlations with Fe or Mn in three or more environments. As expected, PH QTL were located with those of Co, Ca, Na and Zn on chr 1 and 8. In theory, a taller plant will have more biomass and hence is able to accumulate higher levels of Zn during vegetative growth, which then becomes a larger Zn source during seed filling. Lower harvest index of taller plants may also mean that internal competition among grains for the elements, if any, will be weaker compared to the shorter counterparts. Plants can remobilize and move nutrients from vegetative source organs into seeds during the filling of grains45. Hence, the amount of both Fe and Zn in the cereal grain is dependent on the former physiological processes: firstly, their acquisition from the soil by roots, and secondly, transportation to the shoots and further remobilization of stored minerals from leaves when they senesce at grain filling46,47.
Strong positive correlations between Co, Cu, Fe, K, Mg, Mn, and P were observed in three or more environments. This could be explained by an overlap in mechanisms to uptake and transport these elements within the plant. There have been several studies that reported correlations between different trace minerals21,44,48 and between essential minerals and toxic elements21. Genetic mapping has also been attempted to elucidate the genetic basis underlying these correlations in rice and other cereals49–52. Previous studies suggested that gene pleiotropy and QTL co-localization played a role in the correlations among trace minerals21,44,49. Similarly, several correlated traits were associated with the same QTLs either in the same or in a different environment in this study (Table 3). The results confirm that there is a highly complex genetic network controlling grain nutrition levels at multiple loci19,53,54. The co-localisations of Cu- Zn (chr 1), Co-Zn (chr 7), K-P (chr 1), K-Na (chr 2), K-Mn (chr 2) QTL support the possibility of simultaneous improvement of these elements in rice grain. Fe and Zn have been targeted for biofortification for decades55,56 and it is beneficial to explore the possibility to expand to other essential nutrients.
The lack of co-localisation of GY QTL and all grain elements (except for Mo), indicate that selection for higher grain element concentrations at those loci is unlikely to incur yield penalty.
Despite of their strong correlations, P did not share any common QTL with either Zn, Fe or Mg. Thus, the selection for increasing the element concentrations of those at the loci is not likely to increase P concentration, which is desirable in relations to Zn and/or Fe bioavailability. In mature grain, P is mainly stored as phytate (myo-inositol-1,2,3,4,5,6-hexakisphosphate, InsP6), which has the ability to complex Zn and Fe forming insoluble complexes that cannot be digested or absorbed by humans57.
Two genomic regions contained the most QTL for element concentration on chrs 3 and 8 (Table 3). The region around 15.9–16.6 Mb on chr 3 harboured the QTL controlling five elements: B, Ca, Co, Mn and Mo (qB3.1, qCa3.1, qCo3.1, qMn3.1, qMo3.2) and TGW (qTGW3.1). Previous studies have also reported the association of this region with several grain element concentrations including Cd, Cu, Fe, Mn, P and Zn as well as grain length, thousand grain weight, grain yield and heading date20,34,43,51. There has not been any report of the QTL controlling B, Ca or Co concentration in this genomic region to date. On chr 8, the region between 19.6–20.6 Mb harboured the QTL controlling the concentrations of Co, Mg, Na and PH (qCo8.2, qMg8.1, qNa8.1, and qPH8.1). This region was also found to be associated with traits including Cd, Cu and Zn concentrations in the grain, Cu and Mg concentrations in the leaf, photosynthetic ability and plant height in previous studies20,34,51,58,59. This is the first time that a QTL for the grain Co and Na concentrations is being reported in this region. Overall, the two genomic regions on chr 3 and 8 that were associated with multiple elements could lead to the possibility for improvement of multiple nutrients simultaneously in rice breeding. However, grain yield and other developmental traits have also been mapped to the three regions in previous studies, suggesting that selection for higher grain nutrition may incur yield penalty and should be taken into consideration.
For QTLs to be highly effective within breeding programs, they must explain a significant proportion of the variation and be stable across environments and populations31. The stability of the QTL in our study was investigated over four environments. Among the QTL detected for grain elements, qZn7.2 associated with Zn concentration on chr 7 was consistent in all four environments. The QTL with consensus in three environments were qCo7.1 and qK6.1, associated with Co and K concentrations, respectively in three dry growing seasons. The two-environment QTL were found for eight traits including B, Ca, Co, Fe, K, Mo, Na and Zn concentrations. Interestingly, the four-environmental qZn7.2 and the three-environmental qCo7.1 were co-located on chr 7 (~ 29.26 Mb). The QTL accounted for approximately 5–8% and 4–6% of the variation in Zn and Co concentrations, respectively. The alleles associated with increased Zn and Co concentrations were present in less than 20% of the panel accessions indicating this was a rare allele, probably originating from an uncommon genetic pool. Not only was this QTL highly stable in our study, but it has also been identified in different genetic backgrounds. For example, significant QTL for grain Zn and Fe concentrations were reported in this genomic region on chr 7 (~ 29Mb) in a Multi-parent Advanced Generation Intercross (MAGIC) population43 and a mapping population consisting of F6 recombinant inbred lines (RILs) derived from the cross Madhukar × Swarna54. Thus, our results reinforce the significance of the loci in controlling grain Zn density and affirm its potential as a strong target for Zn biofortification. Other traits that have been linked to this genomic region were grain inorganic P concentration50 and heading date58 which may have to be taken into account for breeding purposes.
The three-environment QTL qK6.1 was located on the top of chr 6. This genomic region also harboured QTL for K, Cu and Zn concentrations and heading date in previous studies11,20,60. There has not been any QTL for grain yield reported in either of the genomic regions on chr 6 and 7 indicating that they are promising targets for improving Zn, Co and/or K concentration without yield penalty.
There was no stable QTL detected for grain yield or Cu, Mg, Mn and P concentrations. This is likely attributed to the large effect of the environmental conditions on the traits. Temperature, daylength, rainfall, soil nutrition could affect plant growth, flowering time, grain development, grain yield and grain elements34,61. Although consensus QTL can generally be considered as more favourable for marker-assisted selection, some QTL detected in one environment may lead to important discoveries.
Located within the markers flanking the four-environment QTL for Zn (qZn7.2), there is a prominent potential candidate gene OsNAS3 (Os07g0689600) coding for nicotianamine synthase 3 (Table 4). Nicotianamine synthase (NAS) is the enzyme responsible for production of nicotianamine (NA), a metal chelator for the internal transport of diverse metals, including Cu, Fe, Mn and Zn62. In rice, NA bound to Zn in phloem is supposed to avoid Zn immobilization in the alkaline conditions of the phloem sap, thus playing a vital role in intercellular and long-distance transport of Zn to maintain Zn homeostasis in plants 46,63. Rice possesses three NAS genes, namely OsNAS1-364. Overexpression of each NAS gene led to significant increases of Fe and Zn levels in the rice grain65,66 implying that they can all be targets for improving Zn and Fe concentrations in rice grain. The NAS1 and NAS2 genes, located near each other on chr 3 are highly similar in their sequences and functionality while OsNAS3 (located on chr 7) has a unique expression and likely plays a different role. Under excess Fe/deficient Zn, OsNAS3 may play an important role in maintaining Zn levels in the newest leaves as the NAS3-knockouts plants were impaired in Zn translocation and distribution and tended to decrease Zn levels in new leaves and increase in old leaves67. In addition, OsNAS3 gene may also play a role in mitigating excess Fe as the OsNAS3 expression was as high as that of the other important Fe excess-responsive genes68. No clear functional SNP was identified in OsNAS3, but the presence of SNPs specific to high Zn haplotypes within the OsNAS3 promoter43 suggests that the effect of Zn-7.2 may be achieved through modulating expression of this gene.
Located in close proximity of the qZn7.2 was a gene coding for calmodulin 7 (CAM7) (Os07g0687200). Calmodulin 7 belongs to a Ca2+ ion binding protein family present in the rice phloem sap. This family is considered to play an important role in signal transduction in the sieve tubes of rice plant69. The fact that both NAS3 and CAM7 genes were linked to the stable Zn and Co QTL implies the important roles of the phloem transport processes for Zn and possibly also for Co from vegetative tissues into the grain. In this genomic region, there were also a gene coding for Phytochelatin synthase 12 (Os07g0690800). The enzyme Phytochelatin synthase catalyses the final step in the biosynthesis of phytochelatins, which are shown to be essential for Cd detoxification and Zn tolerance in Arabidopsis thaliana 70,71. The roles of this in regulating grain Zn and/or Co concentrations has not been reported to date.
On chr 3, there were three genes playing key roles in starch synthesis and grain filling including SUS1 (Os03g0401300), GS3 (Os03g0407400) and GS5 (Os03g0393300) (Table 4 & Table S5). SUS1 (linked to qCo3.1) encodes a sucrose synthase (Sucrose-UDP glucosyltransferase) responsible for the biosynthesis of starch within the endosperm. Overexpression of this gene in transgenic rice lines led to increased grain yield (per plant) and TGW72. GS3 (linked to qCa3.2 and qTGW3.1) and GS5 (linked to qMo3.2) encode a transmembrane protein and a putative serine carboxypeptidase, respectively. Natural variations in either of these genes were found to play important roles in regulating grain filling and final grain size and weight73–76. On chr 6, there was one gene coding for granule-bound starch synthase GBSS1 (Os06g0133000) located within the three-environment QTL qK6.1. This enzyme is involved in starch synthesis during grain filling, specifically being responsible for the synthesis of amylose and building the final structure of amylopectin. These results suggest that those genes played significant roles in controlling grain filling and TGW, which in turn affected grain nutrient concentrations in our study, particularly for Ca, Co, K and Mo. The transfer route of micronutrients (such as Fe and Zn) into the grain is thought to be similar to that of sucrose77,78. In transgenic wheat lines overexpressing a sucrose transporter gene, there was an increase in grain yield as well as a 20–40% increase in grain Fe and Zn concentrations 78. The functionality of those genes in relation to controlling grain nutrient elements such as Ca, Co, K and Mo in rice, will require further studies to elucidate.
Underlying the QTL clusters, there were also several genes that play important roles in regulating flowering time and whole-plant senescence in rice including Terminal flower 1-like protein, FT7-Flowering Locus and NAC factor (Table 4, Table S5). Flowering, grain filling and whole-plant senescence are processes that are highly important in determining grain weight, yield and quality parameters such as grain protein content (GPC) and grain micronutrient including Fe, Mn and Zn levels in cereals79,80.
There are several transporter genes linked to the stable QTL for K concentration qK6.1 and the QTL clusters on chrs 3 and 8. On chr 3, two metal transporters; i.e OsZIP2 (Os03g0411800) and a heavy metal transport/detoxification (Os03g0412300) were located within the QTL for Ca concentration and TGW. On chr 6, a high-affinity potassium transporter genes (OsHKT9) were linked to qK6.1. On chr 8, a Cation efflux protein MTP12 (Metal tolerance protein) was linked to qPH8.3. Amongst these genes, the functionality of the ZIP transporter families has been well characterised. There are 17 ZIP transporters in rice and most of them show broad substrate transport activity (transporting Zn, Fe, Mn, Cd and Co). The genes OsZIP1, 2, 4, 5, 7, and 8 are highly induced by Zn deficiency81–83. It has been shown that mineral uptake and transportation in rice is a complex process that involved the combined actions of several transporter genes45,84. The genes being identified in this study would be potential candidates for further studies to improve essential nutrient element concentration in the rice grain.
In conclusion, the rice diversity panel used in this study proved to be a useful resource for association mapping of rice grain nutrition with significant variation observed and QTL detected for all traits. Co-localizations of QTL for multiple grain element concentrations was found, and particularly those on chrs 3 and 8 open the opportunity for enhancing multi elements simultaneously. Consistent QTL across environments were identified, particularly the four-environment QTL for Zn (qZn7.2). This QTL had been reported previously, indicating its stability in different genetic backgrounds, and is a strong candidate for being used in breeding for higher Zn concentration. Multiple candidate genes were identified, which play various roles in controlling mineral accumulations in rice grain including NAS3 and SUS1. Further gene functionality studies would be helpful to validate the significance of the candidate genes in breeding for higher micronutrient content in rice grains.