Identification of genomic regions associated with agronomic and biofortification traits in DH populations of rice

Rice provides energy and nutrition to more than half of the world’s population. Breeding rice varieties with the increased levels of bioavailable micronutrients is one of the most sustainable approaches to tackle micronutrient malnutrition. So, high zinc and iron content in the grain are primary targets in rice biofortification breeding. In this study, we conducted QTL mapping using doubled haploid (DH) populations, PSBRc82 x Joryeongbyeo and PSBRc82 x IR69428, phenotyped for agronomic traits and micronutrients during two growing seasons and using genotypic information from analysis with the 6K SNP chip. A number of DH lines were identified as having high grain Zn and Fe content in polished rice. Importantly, we identified 20 QTLs for agronomic traits and 59 QTLs for a number of biofortification traits. Of the 79 QTLs, 12 were large-effect QTLs (>25% PVE), nine QTLs were consistent across seasons in either population, and one QTL was identified in both populations. Moreover, at least two QTLs were clustered in defined regions of chromosomes 1, 2, 3, 4, 5, 7 and 9. Eight epistatic interactions were detected for Cu, Mg, Na, and Zn in population 1. Furthermore, we identified several candidate genes near QTLs for grain Zn (OsNRAMP, OsNAS, OsZIP, OsYSL, OsFER, and OsZIFL family) and grain yield (OsSPL14 and OsSPL16). These new QTLs and candidate genes help to further elucidate the genetic basis for grain micronutrient concentration, and may prove useful for marker assisted breeding for this important trait.


Introduction
Rice (Oryza sativa L.) is one of the most vital food plants in the world. It provides energy and nutrition to nearly half of the people on earth [1]. In most of the developing countries in Asia, rice is eaten in significant quantities almost every day and it is the major component of the daily diet of the population since there are limited opportunities and resources to diversify the diet with fruits, vegetables, and meat [2]. Further, modern high-yielding rice varieties are low in mineral elements with polished rice having even lower amounts. Thus, milled rice is not a major supplier of any of the mineral elements in significant quantities and cannot meet the recommended daily dietary intake [3]. Consequently, most rice-eating, resource-poor people in South and Southeast Asia, Africa, and Latin America suffer from chronic micronutrient malnutrition, often called hidden hunger [4].
Mineral elements, such as iron (Fe), zinc (Zn), calcium (Ca), magnesium (Mg), potassium (K), manganese (Mn), boron (B), phosphorus (P), and copper (Cu), are essential for many cellular and metabolic functions and important structural components of many tissues, fluids and vital organs. So, they are highly beneficial for the growth and development of plants and animals [5]. However, some elements, such as arsenic (As) and cadmium (Cd), have detrimental effects on humans [6]. Arsenic causes cancer, neurological disorders, and circulatory diseases; Cd causes renal and liver disorders, weak bones, anemia, and hypertension [7]. Thus, the concentrations of these mineral elements in staple foods, such as rice, have a huge implication on human health. In addition, the potential contamination of rice with toxic elements may also have a huge impact on the rice trade [8].
Among different mineral elements, Fe and Zn are essential for human health. Zn is a major cofactor for several vital enzymes and Fe is an important component of blood. Deficiency of these elements in the diet results serious human health problems, especially among children and pregnant and lactating women [9]. Zn deficiency causes stunting, diarrhea, and reduced immunity, while Fe deficiency causes anemia. Approximately 2 billion people suffer from Fe and Zn deficiencies globally, and it is a high priority to address these deficiencies in order to achieve the sustainable development goals such as reduced child and maternal mortalities and a reduction in poverty and hunger [10]. Biofortification of staple crops has emerged as a sustainable strategy to overcome mineral deficiencies. Using this approach, CGIAR centers in collaboration with HarvestPlus have made significant progress in developing, releasing, and disseminating biofortified crops [4].
The accumulation of mineral elements in the grain is a complex process and highly influenced by environmental factors. This made early-generation phenotypic-based selections for grain mineral elements, particularly grain Zn less effective and has slowed progress in breeding for biofortified rice varieties [11]. A thorough understanding of the genetic basis of grain mineral elements at the molecular level and identifying major-effect QTLs can assist in faster development of biofortified rice varieties through marker-assisted breeding (MAB) [12].
As rice has a relatively small genome and has long been the focus of significant genetic research as a model for the cereal crops, there are enormous genomic resources available, including genome-wide single-nucleotide polymorphic (SNP) molecular markers and various advanced genomic platforms, to allow dissection of complex traits at the molecular level [13]. Some of the recent efforts to map QTLs for mineral nutrients include the use of introgression lines (ILs) for Mn, Ca, Cu, Mg, P, K, Fe, and Zn [3] and DHs to discover QTLs for P, Cu, Mn, Fe, and Zn [14]. Several major-effect epistatic loci were reported for Fe and Zn [15]. Twentythree and nine QTLs, respectively, were identified in two environments for grain concentrations of Ca, Fe, K, Mg, Mn, P, and Zn [16]. Using RILs, 14 QTLs for grain Fe and Zn were identified and prioritized the candidate genes underlying these QTLs and known to be involved in metal homeostasis: OsYSL1 and OsMTP1 for Fe; OsNAS1 and OsNAS2 for Zn; and OsNAS3, OsNRAMP1, and APRT for both Fe and Zn [17]. Similarly, several metal homeostasis genes validated in a RIL population shown significant association with grain Zn content [18]. Meanwhile, QTLs were mapped for 19 mineral elements under multiple environments [12]. All these studies resulted in identification of multiple loci distributed throughout the genome with low to moderate genetic effects [12]. The majority of these loci were specific to different genetic backgrounds and environments and seldom used in MAB.
Among different biparental populations, DHs are fixed genetic materials that can be developed faster relative to other mapping populations, can be evaluated across years and locations readily, and have less genetic background 'noise' that make them important genetic resources for mapping QTLs/genes for various traits [19]. Several reports have shown the utility of DH populations in identifying QTLs for the concentration of elements in grain [14] [20].
The main objectives of our study were to evaluate doubled haploid mapping populations for yield, yield components, and grain micronutrient traits; map QTLs for yield and grain micronutrients; understand the QTL x QTL interactions; identify candidate genes associated with major effect Zn QTLs; compare QTLs for grain micronutrients with earlier QTLs; and identify promising lines with high grain Zn and yield.

Experimental location and plant materials
These studies were conducted during the 2015DS (dry season) and 2015WS (wet season) at the International Rice Research Institute (IRRI), Philippines. We used two DH populations consisting, respectively, of 130 and 97 lines derived from PSBRc82 x Joryeongbyeo (P1) and PSBRc82 x IR69428 (P2). The PSBRc82 is one of the most widely adapted and popular rice varieties in the Philippines, while Joryeongbyeo is a Korean rice variety found to have high Zn content and IR69428 is a breeding line (IR 65564-44-5-2/SENGKEU//IR 65600-1-3-2) having high Zn content.

Phenotypic evaluation of mapping populations
The experiments were carried out following a randomized complete block design (RCBD) with three replications; with check varieties used for yield and micronutrient comparisons. The populations were phenotyped for four agronomic traits and 13 grain micronutrient traits. The agronomic traits were measured following the standard evaluation system [21], including days to 50% flowering (DTF; number of days taken from sowing to the time that 50% of plants of the family showed flowering), plant height (PH; height from the soil surface to the tip of the primary panicle identified at the time of harvest), number of tillers (NT; average number of tillers from three plants at the time of harvest) and yield per hectare (YLD; average weight of the cleaned grains dried to 14% moisture from all the plants per plot). For mineral analysis of the grain, 50 g of paddy samples were dehulled using a Satake dehuller and milled for one minute using a K-710 mini-lab rice polisher. Milled rice samples weighing at least 3g representing each plot were analyzed using X-ray Fluorescence Spectrometry (XRF) (Oxford) to measure Fe and Zn [22]. While, brown rice samples from two replications were analyzed for all the biofortification elements using Inductively Coupled Plasma Mass Spectrometry (ICP-MS) at Flinders University, Australia. The average reading per plot was used for subsequent statistical analysis.

Genotyping
We collected leaf samples from 227 DH lines and parental lines, which were ground after freezing with liquid nitrogen. The genomic DNA was extracted using modified CTAB protocol [23]. The DNA quality was checked by 1.5% of Agarose gel electrophoresis. High-quality DNA samples with appropriate concentrations (~50 ng) were submitted for 6K SNP Infinium Assay. Scanned image calls and automatic allele calling were loaded in the Illumina Genome Studio data analysis version V2010.1.

Statistical analysis
Descriptive statistics were generated using STAR v.2.0.1, while Analysis of Variance (ANOVA) was carried out using PBTools v1.4. Histograms and correlations between pairs of traits were estimated through Pearson correlation co-efficients using R software. The model used for ANOVA was: where μ is the overall mean, α i is the effect of the i th genotype; r j is the effect of the j th replicate, b kj the effect of the k th block within the j th replicate and ε ijk the error. The genotypes were considered fixed while replicates and blocks within replicates were random.
For estimating broad sense heritability, variance components were estimated considering all factors including genotypes as random. For each group, broad-sense heritability or repeatability (H 2 ) for each season was calculated as: Where s 2 p is the phenotypic variance, s 2 g is the genotypic variance, s 2 e is the error variance and r is the number of replications in the season.
Linkage mapping and QTL analysis. Marker data sets generated from 6K SNP genotyping were screened based on >80% call rate, homozygosity, and polymorphism between respective parents. Scoring of alleles in DH populations was adjusted at each SNP locus through comparing parental alleles at the respective SNP locus as dominant markers. The maternal alleles were scored as 'A', paternal alleles as 'B'. Ambiguous or missing data were shown as "X". Redundant SNP markers (i.e., completely correlated markers) were removed and the remaining SNP markers were anchored, grouped, and ordered by the input. Moreover, SNP markers that were considered unlinked were removed from the linkage map in order to attain the appropriate linkage map lengths. Linkage maps of the two DH populations were created following Kosambi function using IciMapping ver.4.1 [24]. QTLs were identified by inclusive composite interval mapping (ICIM) in QTL IciMapping ver.4.1. The average trait values for each line in each DH population were used for the QTL analysis. QTLs were named following the standard procedures. The LOD thresholds of the QTL were set based on 1,000-permutation test at a 95% confidence level. The proportion of observed phenotypic variance explained by each QTL and the corresponding additive effects were also estimated. The digenic (epistatic) interactions between marker loci were determined by setting the LOD threshold of 6.0. QTLs were visualized using MapChart v.2.3 [25].

QTL search and candidate gene analysis
Annotated genes with functions related to metal homeostasis, loading, sequestration, and transport of mineral elements were compiled. The physical positions of these annotated genes were determined using the RAP-DB Genome Browser (http://rapdb.dna.affrc.go.jp/viewer/ gbrowse/irgsp1). Annotation and functions attributed to different candidate genes were downloaded from Oryzabase (https://shigen.nig.ac.jp/rice/oryzabase/gene) and RiceChip (www. ricechip.org). Metal transport and heavy metal homeostasis genes physically located within or near QTLs for agronomic and biofortification traits were considered candidate genes.

Phenotypic analysis
All the agronomic and biofortification traits showed typical normal distributions (Table 1, S1 and S2 Figs). The analysis of variance showed that there was significant genotypic variation for all the traits in the two populations during the DS and WS. The mean, range, standard derivation (SD), co-efficient of variation (CV), and heritability (H 2 ) values are presented in Table 1.
The highest mean values for PH, DTF, and NT were recorded in WS, whereas the highest value of YLD was observed during DS. The general trend of biofortification traits was that higher values were recorded during DS except for grain Fe and Zn. The CV was lower than 10% for DTF, K, Mg, and P in P1 and for Mg, and P in P2, whereas CVs ranged from moderate to high (10.38%-52.61%) for all the other traits. Broad sense heritability was high (60%-97%) for all the traits in both the seasons and populations, except for Fe and NT.
Pearson correlation results indicated that the number of significant correlations between traits was higher in the DS compared to the WS (Figs 1 and 2). Similarly, the number of positive correlations was higher in P1. Notable significant positive correlations were observed among phenotypic traits; including PH and YLD, and NT and YLD, among biofortification traits between Zn and Fe; while highly significant negative correlation was observed between YLD and Zn, and YLD and Fe. Furthermore, positive and negative significant correlations were also observed in other mineral elements. For example, Fe and Zn were positively correlated with other mineral elements such as Mg, Co, Cu, K, Ca, P, B, Mo, and Mn except Na and As, while negative correlations were found among NT, PH, DTF, and YLD.

SNP-based molecular map construction
Of the 6,404 SNP markers used to genotype the parents and progenies of each population, 38.2% were polymorphic between parents of P1; while 33.2% were polymorphic between parents of P2. Removal of redundant markers resulted in 469 SNP markers and 398 SNP markers remaining for linkage map construction for P1 and P2, respectively. The distribution of SNP markers varied across different chromosomes (S1 Table). The highest and the lowest numbers of SNPs were on chromosomes 2 (65) and 12 (10), respectively in P1, while the highest and the lowest numbers of SNPs were on chromosomes 3 (86) and 8 (10), respectively in P2. The total lengths of the linkage maps of P1 and P2 were 1396.5 and 1619.2 cM, respectively. The average marker intervals were 2.98cM and 4.07 cM in P1 and P2, respectively.

QTL analysis
QTL analysis identified 79 QTLs for 16 traits during 2015DS and WS in both P1 and P2 (Table 2 and Fig 3). Most of the QTLs contributed more than 10% of the total phenotypic variance explained (PVE) and also had a high additive effect on the traits. The details of the QTLs identified are provided in Table 2 YLD: Seven QTLs on chromosomes 1, 2, 4, 8, 9, and 11 were identified for grain yield in P1 and P2 during two seasons. These QTLs have PVE that ranged from 5.2% to 27.5% and were derived from high-yielding parent except qYLD 1.1 . Further, all QTLs for yield were identified in the WS except qYLD 2.1 .   QTLs that are consistent across seasons and those with large effects (PVE>25%) could be targeted for marker-assisted selection. We have identified 79 QTLs for agronomic and biofortification traits. Of these, six QTLs were consistent across seasons in P1 (qPH 4.1 , qCa 2.1 , qMn 2.1 , qMo 1.1 , qMo 1.3 , and qZn 2.1 ), three were detected across seasons in P2 (qDTF 1.2 , qDTF 7.1 , qDTF 10.1 ). The QTL qMo 12.1 was identified in both populations and was consistent across seasons in P2. Further, 12 QTLs have large effects: qYLD 4.1 , qAs 1.1 , qAs 10.1 , qCa 2.1 ,  qCo 1.1 , qCo 4.1 , qCu 4.1 , qMn 2.1 , qMo 1.1 , qMo 1.2 , qNa 1.1 , and qNa 10.1 . It is also noteworthy that qCa 2.1 , qMn 2.1 , and qMo 1.1 are large-effect QTLs detected in P1 and consistent across seasons.
Co-location of QTLs for agronomic and biofortification traits. In all, fourteen QTL clusters each consisting of two to six QTLs were identified (S2 Table and Fig 3). Four major QTL clusters consisting of at least four QTLs were observed on chromosomes 1, 2, and 4. The first of which, located between 23.08 Mb to 23.7 Mb on chromosome 1 and it included four QTLs, qDTF 1 9.1 and qK 9.1 , and qMg 9.1 on chromosome 9 were found to be co-located.
A number of correlated traits share common QTLs detected in the same population and season. The negatively correlated YLD and P share common QTLs on chromosomes 1 and 2 detected in P1 during the WS. The positively correlated K and P have common QTLs on chromosome 2 detected in P1 during the DS and on chromosome 5 detected in the same population during the WS. Both K and P are negatively correlated with YLD; these three traits share a common QTL on chromosome 2 detected in P1 during the DS. The positively correlated Ca, Mn, and Zn share a common QTL on chromosome 2 identified in P1 during the WS. The positively correlated PH and YLD were both negatively correlated to K, these three sharing a common QTL on chromosome 4 identified in P1 during the WS. The negatively correlated As and Mg share a common QTL on chromosome 5 detected in P1 during the WS. The negatively correlated DTF and PH share a common QTL on chromosome 7 uncovered in P2 during DS. The negatively correlated YLD and K share a common QTL on chromosome 9 detected in P2 during WS.

Epistasis for agronomic and biofortification traits
Eight di-genic interactions were detected in P2 for Cu, Mg, Na, and Zn while no interactions were observed in P1 (Table 3). Three epistatic interactions between loci on chromosomes 3, 8, and 10 for Na explained~28.6% PVE each. One epistatic interaction was identified for Mg between loci on chromosomes 1 and 7 that contributed 34.1% of the PVE. Two epistatic interactions for Cu, detected on chromosome 5 and between chromosomes 8 and 9, accounted for 43.2 and 28.1% of the PVE, respectively. For Zn, 2 epistatic interactions were identified that contributed 16.9 and 32.6% of the PVE, one between loci on chromosomes 5 and 9 and the other between loci on chromosomes 3 and 5.

Candidate genes located near associated SNPs
Candidate genes located within and near (<3.0 Mb) 43 QTLs of agronomic and biofortification traits were identified (S3 Table). For DTF, six main candidate genes OsCCT01, HESO1 homolog, OsLAX1, OsHAP3E, OscpSRP43 and Ef2 were identified. Meanwhile, OsSPL16 was main candidate gene, identified for grain yield.
For micronutrient traits, the primary candidate genes are those involved in metal and cation transport and those with transferase activity such as OsALA2, OsAHA8, and OsECA1 for Ca; OsMTP6, SOD, and OsNRAMP7 for Co; OsSTA107 and OsCOPT4 for Cu; OsHAK1, OsGLR3.1, OsHKT4, OsHKT7, OsHKT15, OsHAK11, and OsHAK18 for K; OsMRS2-5 and OsMRS2-7 for Mg; OsNRAMP1 and OsNRAMP5 for Mn; OsEFCAX1 and OsHKT8 for Na; OsPAP9b, OsABC1-2, and OsPPDKB for P. One candidate gene OsACR2.1 is involved in As metabolism. Other candidate genes are involved in response and sensitivity to micronutrients such as OsDJ-1A, OsDJ-1B, and several OsSultr genes for As; OsCDT1 and OsCBSCLC4 for Cu; and OsOCP for Fe. Others are involved in calcium-mediated signaling and detection such as OsCam1-3, OsCam3, and OsCBL5.
For Zn, the primary candidate genes were OsMHX1, OsYSL2, OsMTP6, OsMTP12, OsTOM1, OsTOM2, OsZIFL7, and OsZIFL8, and their functions are mainly for Zn ion transport as well as OsNAS1 and OsNAS2 that are involved in mediating the uptake of Zn from the soil. Other candidate genes are involved in response and sensitivity to the Zn ion such as OsP-DIL1-3, OsYSL15, and OsFER1.

Breeding lines with high yield and grain Zn
Eight breeding lines (Four lines per population) were identified as having high yield and high grain Zn based on their field performance ( Table 4). All identified lines were comparable with IR64 in terms of yield for both seasons. Meanwhile, six of the eight lines consistently exhibited comparable Zn with check IR69428 during both DS and WS while two lines were only comparable with the check IR69428 during WS. Upon closer examination, each of the eight lines with high grain Zn and yield harbored one to three Zn QTLs, which account for from 14 to 50% of the Zn QTLs identified for each respective population. Of the seven Zn QTLs uncovered in P1, three were identified in IR91143-AC 89-1 while the other three lines either have one or two Zn QTLs. Meanwhile, of the two Zn QTLs uncovered in P2, only one was identified in the four lines.

Discussion
Biofortification is considered the most promising, food-based approach to address micronutrient malnutrition [26]. There have been significant efforts over the last decade to biofortify the major cereals, pulses, and tuber crops targeted to different parts of the world and a lot of progress has already been achieved in this endeavor [8]. Rice, the dominant cereal and a major staple in Asia, is the prime target for biofortification to tackle micronutrient malnutrition in South Asia and Southeast Asia [27]. In order to make progress in biofortification of rice, there is a requirement for enough variability in grain micronutrients in relevant germplasm collections, a better understanding of the genetic basis of the grain micronutrients, and their interactions with genetic backgrounds and environmental factors [12]. Recent advances in rice genomics have enabled breeders to dissect the molecular basis of complex grain micronutrient traits to identify QTLs/genes for use in MAB [28], [29], [30]. At IRRI, we developed and evaluated two DH populations and mapped QTLs for agronomic and grain micronutrients. The analysis of variance indicated the significant genotype effects for most of the traits. The mean, range and CV also indicated wide variation for all the agronomic and grain micronutrients traits in both the populations and seasons. Also, most of the traits showed normal distributions indicating a complex genetic basis. It is a well-known fact that polygenically inherited complex traits, such as agronomic traits, yield, and grain micronutrient traits, show wide variation and are greatly affected by GxE [17]. Several earlier studies on analysis of multiple micronutrients in rice grains from germplasm collections and mapping populations evaluated in different environments provided similar results [3], [15], [28], [31], [32].
Moderate to high broad-sense heritability (H 2 >60%) was observed for most of the traits except for NT and Fe, suggesting that they are genetically controlled and amenable to genetic manipulation. A high H 2 (>50%) and significant GxE for mineral elements, such as Zn, Cu, Mo, and Mg, observed across the populations and locations in most the recent studies on grain micronutrients [12], [17], [28]. The association results clearly show that there are positive correlations among different micronutrients, but they have a negative association with YLD; while Fe and Zn were strongly positively correlated irrespective of the seasons, locations, and populations. Several studies have reported the positive relationship between Fe and Zn [14], [33] as well as the negative relationship between YLD and Zn in rice [28], [34], [35]. However, there are some reports showing positive or no significant correlations between yield and Zn [36], [37]. A positive correlation between Fe and Zn could allow simultaneous improvement of both minerals [22]. However, the negative linkages between YLD and Zn must be eliminated for successful biofortification. Hence it is necessary to identify high-Zn parental lines with acceptable yield potential, by designing appropriate breeding strategies, selection schemes and evaluation procedures for the successful development and release of high-Zn rice varieties [38], [39]. The SNP marker analysis revealed high rates of polymorphism (>30%) between the parents, and this was expected because the paternal parents IR69428 (japonica derivative) and Joryeongbyeo (japonica) were genetically distant from PSBRc82 (a popular indica rice variety). Eventhough polymorphism was higher but the final set of markers used for linkage map construction were 469 SNP and 398 SNPs in the P1 and P2, this is mainly because we removed markers with ambiguous or missing allelic calls, redundant, unanchored and unlinked SNP markers. All these steps resulted in less number markers on some chromosomes leading to some gaps. However, over all the SNP set gave good genome coverage with a marker density of 2.98cM and 4.07 cM which was enough for the linkage map consutruction and QTL analysis in a DH population.
We detected 20 QTLs for agronomic traits and 59 QTLs for mineral elements, almost 40% of them were derived from the paternal parents. The most significant QTLs identified in our study were qAs 1.1 , qAs 10 10.1 and each had a PVE of more than 25%. Allelic effects of the QTLs for Cu, K, Mn, and P were from either of the paternal parents (IR69428 or Joryeongbyeo), while Cu and P QTLs were specific to Joryeongbyeo. In general, most of the QTLs for YLD and PH were contributed by PSBRc82, while QTLs for mineral elements were contributed by the paternal parents. But only 10 QTLs were consistent across the seasons and most of them, such as qCa 2.1 , qMn 2.1 , qMo 1.2 , Mo 1.3 , qMo 12.1 , and qZn 2.1 , were detected only in P1. It is notable that qMo 12.1 was the only QTL consistent across the populations and seasons, indicating the significant genetic background effect and significant GxE for all the other mineral elements. For YLD, all the QTLs except qYLD 1.1 were contributed by PSBRc82, with a PVE of 5% and additive effect of 0.24 t ha -1 . Conversely, 6 of the 8 QTLs for grain Zn were derived from paternal parents; all of them were season-and population-specific except qZn 2.1 . Recent studies on QTL mapping for mineral elements in rice and other crops using RIL, ILs, F 2 , DH, and MAGIC populations have identified multiple loci and clearly demonstrated the genetic complexity of the grain micronutrient traits [3], [14], [17]. Some of the loci identified in our study corresponded to the same chromosomal regions and traits as for other reports. However, thorough characterization and validation of the QTLs/genes are prerequisites before pursuing MAS. Since most of the QTLs are season and genetic background-specific, it will be necessary to pool multiple loci so that marker-assisted QTL pyramiding, marker-aided recurrent selection, or genomic selection can be appropriate strategies to develop rice varieties with improved micronutrient content.
Different mineral elements may share the same or similar pathways for their uptake, transport, and loading, consequently they may also share same genomic regions and QTL/genes. We identified several QTL clusters for multiple mineral elements and also for the agronomic traits. Present findings detected 14 co-localized QTLs responsible for different traits at specific genomic regions across chromosomes 1, 2, 3, 4, 5, 7, 9, and 12. Eleven of 14 co-localized QTLs were identified in both the WS and DS. Moreover, 4 of 11 were not only detected in both seasons, but also in both populations. Such trait co-locations have also been reported for Mg, Cu, Si, Se, Fe, K, Mn, and P [16], [28], [40]. Similarly, several clusters of QTLs were also found to be involved with grain minerals such as Zn, Fe, Mn, Zn/Fe, Mg, and Cu [41]. If the linked traits have positive correlations they can be simultaneously improved, however the negative linkages must be broken before their use in breeding.
Several epistatic QTLs were found for Cu, Mg, Na, and Zn with large effect QTLs contributing 16.9 to 43.2% of the PVE. Six epistatic QTLs identified for grain Zn accounted for 50.2% of the heritability of the trait [15]. Epistatic QTL for grain Zn with 20% of PVE was identified, which indicated strong genetic control involving many QTLs or genes [28]. Another epistatic QTL for Zn concentration was also observed explaining 60% of the total variance with an additive effect of 11.26 ppm [17]. A possible explanation for this observation may involve genetic changes wherein gene by gene interactions could modify the expression of phenotype and physiology [42]. Thus, they play a potential role in controlling grain mineral elements in polished rice.
The candidate gene analysis revealed that 16 known genes for agronomic traits and 52 candidate genes involved in metal homeostasis were found to co-locate with different QTLs identified in our study, emphasizing the importance of these loci for further molecular characterization or use in breeding. Some of the important gene families harboring the major QTLs are OsNRAMP, OsZIP OsYSL, OsNAS, OsFER, OsZIFL, and OsOCP. Their roles in metal uptake, transport, and loading are well characterized and their over-expression has shown a several-fold increase in mineral accumulation in different plant parts of rice [43][44][45][46][47][48][49]. Similarly, for agronomic traits, several important genes, such as OsSPL14 and OsSPL16, were located near qYLD 8.1 . OsSPL14, which promotes panicle branching and higher grain productivity, might be useful for increasing grain yield in rice [50]. The expression of the OsSPL16 gene has been shown to be involved in control of grain size, shape, and quality in rice [51].
The present study showed that several breeding DH lines with high grain Zn and grain yield were significantly high in both the populations as well as across the seasons. However, several previous studies have reported a significant negative relationship between grain Zn and grain yield in rice [28], [34], [35]. We identified eight DH lines with Zn more than 18 ppm with good agronomic traits. It is interesting note that among the DH lines IR91143-AC 122-3 showed yield and Zn content comparable to the parental line IR69428 but it had only two of the eight Zn QTLs. This may be due to the genetic background effect or transgressive variation due to epitasis and can be dissected further to understand the underlying mechanisms. These DH lines can be used as donors in breeding programs or can be directly tested in multi-location trials to further evaluate their performance.

Conclusions
Considerable genetic variation for all traits was observed in the two DH populations. Seventynine QTLs were detected for agronomic and biofortification traits through inclusive composite interval mapping. The majority of the QTLs accounted for greater than 10% of the PVE. Grain Zn QTLs was found to cluster with QTLs for other biofortification traits, such as B, Ca, Mn, Mg, As, K, and Co. Six epistatic interactions accounted for a larger proportions of PVE ranging from 16.89 to 43.22%. Importantly, several candidate genes, such as OsNRAMP, OsNAS, OsZIP, OsYSL, OsFER, and OsZIFL were identified as necessary for grain Fe and Zn accumulation, together with OsSPL14, and OsSPL16 for increasing grain yield.