In this study, 617 Indica hybrid varieties were collected from hybrid rice-growing regions in southern China to investigate grain Cd accumulation. The majority of them were elite commercial hybrids that are presently grown. All of the hybrids were planted in two different Cd-polluted paddy fields (Beishan, China) in the 2014 rice-growing season in the experimental fields of the Institute of Subtropical Agriculture of the Chinese Academy of Sciences (ISA) in Changsha, China. The soil Cd concentration was 1.80 mg/kg in field I and 2.45 mg/kg in field II, with pH values of 5.4 and 5.5, respectively. To determine the differences in grain Cd levels between inbred rice cultivars and hybrids, a total of 68 inbred rice (Oryza sativa) varieties, obtained from the International Rice Research Institute (IRRI), were used. All varieties can be clustered into two subspecies, such as Indica and Japonica, and five ecotypes, such as indica (ind), aus, aromatic (aro), temperate japonica (tej), and tropical japonica (trj) (Garris et al., 2005). The rice cultivars were planted in three different Cd-polluted paddy fields (Zhuzhou and Beishan) from 2012 to 2014. The soil Cd contents of the three fields were 2.80, 1.90, and 1.80 mg/kg, with pH values of 5.2, 5.0, and 5.4, respectively. The rice hybrids and cultivars were planted in the same fields in Beishan in 2014.

The experiments were arranged in a randomized complete block design with three replicates with a spacing of 17 cm between plants and a distance of 20 cm between rows in each field. Field management followed normal agricultural practices except that intermittent irrigation was adopted to maximize the phenotypic differences.

Rice grain was harvested and air-dried to reduce the water content. After milling the grains into brown rice, samples of 200.0 g were ground into powder and oven-dried at 80⋅C to constant weight for determination of Cd concentrations. For wet digestions, 2.0000 g subsamples were placed in an acid mixture of HNO₃-HClO₄ (6:1, v/v). The process of wet digestion was similar to that in our previous study (Yao et al., 2015). Blank digests with no samples and the certified standard material samples (CRM rice; GBW10045, GSB-23) were also treated the same way. The concentrations of Cd and four other metal elements, i.e. Fe, Zn, Mn, and Cu, were determined on an inductive coupled plasma emission spectrometer (Agilent Technologies 700 Series ICP-OES, USA), using the operating parameters as recommended by the manufacturer. Certified standard material (rice flour) was used to ensure the precision of the elemental analysis. Each sample was run in duplicate to ensure the repeatability of the results.

To investigate the genetic background of hybrids, a population structure analysis was performed. We used a total of 36 markers, including 22 simple sequence repeat (SSR) and 14 insertion-deletion (In/Del) markers; the marker loci are evenly distributed on the 12 chromosomes and were selected based on the rice genetic map (Temnykh et al., 2001). The sequences of these marker primers were designed based on the Gramene data base (http://archive.gramene.org/; Supplemental Table 3).

The genomic DNA from each hybrid was extracted using a modified potassium acetate-SDS protocol (Dellaporta et al., 1983). PCR amplifications were performed using standard protocols with minor modifications. The PCR products of the SSR markers were separated and sized by capillary electrophoresis using the DNA analyzer (Fragment Analyzer INFINITY™, Advanced Analytical Technologies, Inc., USA). The products of In/Del markers were separated and visualized on 4% polyacrylamide denaturing gels or 4% agarose gels to distinguish allelic polymorphism.

STUCTURE 2.3.4 was used to evaluate the population structure with a maximum likelihood method by a burn-in of 10,000, a run length of 100,000, and a model allowing for admixture and correlated allele frequencies (Pritchard et al., 2000). Three independent runs yielded consistent likelihoods of the population structure for each predicted K value. The neighbor-joining (NJ) tree was constructed based on a distance matrix calculated by PowerMarker 3.25 (Liu and Muse, 2005). MEGA 5.0 (Tamura et al., 2011) was used to generate the phylogenetic tree.

Based on previous reports (Ueno et al., 2009, 2010; Ishikawa et al., 2010; Yan et al., 2013; Zhang et al., 2013), a total of 14 QTLs (24 linkage markers) were utilized and genotyped in hybrid rice (Supplemental Table 4). The Q matrix was obtained with STRUCTURE (version 2.3.4) and the Kinship matrix was obtained with SPAGeDi 1.5 (Hardy and Vekemans, 2002). Based on inferred ancestry of the hybrids (Q matrix) and the relative kinship (K matrix), a general linear model (GLM) and a mixed linear model (MLM) were used to detect associations between QTL-markers and grain Cd concentrations using the TASSEL program (version 2.0; Bradbury et al., 2007). A significance threshold of 0.01 was used.

Basic statistical analyses, such as Student's t-test, linear regression and two-way ANOVA, were performed mainly with Microsoft Excel 2010. Based on the ANOVA results, the phenotypic variances were partitioned and different components were calculated. Hence, genotypic variations, soil Cd levels, and the interaction between them could be estimated. The proportion of each variation in phenotypic variations could describe the contribution (Var%) of genotype, soil Cd concentration, or their interaction. Two significance levels were applied (P < 0.01 and P < 0.05).

In this study, a total of 617 elite hybrid rice varieties collected from the middle and lower Yangtze River Valley in China were analyzed for their performance with respect to grain Cd accumulation. All of the hybrid varieties were planted for phenotyping the grain Cd concentrations in two paddy fields in 2014. The soil Cd concentrations were 1.80 mg/kg and 2.45 mg/kg in field I and field II with pH values of 5.4 and 5.8, respectively. As a quantitative trait, grain Cd concentration in the hybrids varied greatly (Figure 1A). The levels of Cd in the grain followed a normal distribution in both fields. In field I, grain Cd levels ranged from 0.67 to 6.58 mg/kg, with an average of 3.16 mg/kg and a median of 3.22 mg/kg. In field II, Cd levels ranged from 0.91 to 7.83 mg/kg, with an average of 3.68 mg/kg and a median of 3.63 mg/kg. In most cases, a cultivar grown in field II accumulated more Cd in the grain than it did in field I (R² = 0.72, P = 4.0E-88), indicating that the grain Cd concentrations were greatly affected by the soil Cd levels (Figure 1C). The large variations in grain Cd accumulation observed in the different genotypes in the two fields suggested that genetic factors contributes to phenotypic divergence. To estimate the effects of genetic components in determining grain Cd accumulation, the phenotypic variations were divided into genotypic variations, soil Cd levels, and their interactions. By two-way ANOVA, all three parts had significant effects in determining grain Cd accumulations (P < 0.01; Supplemental Table 1). The heritability of grain Cd levels was detected as 50.8%, which predicted a stable heritability for grain Cd accumulation in hybrid rice.

We also analyzed the grain Cd performance in inbred rice cultivars independently. Sixty-eight cultivars were planted in three Cd polluted fields with varying Cd levels in three successive years (from 2012 to 2014; Figures 1B,D). The soil Cd contents of the three fields were 2.80, 1.90, and 1.80 mg/kg with pH values of 5.2, 5.0, and 5.4, respectively. Correspondingly, the average grain Cd concentrations for rice grown in these fields were 2.64, 2.29, and 2.43 mg/kg. Similar to the hybrids, genetic variations, soil-Cd levels, and their interactions are also key factors in determining phenotypic variation in the inbred cultivars (P < 0.01; Supplemental Table 1). The 68 cultivars were divided into two subspecies and five ecotypes; e.g. ind, aus, aro, tej, and trj, of which the first three belong to the Indica subspecies while the latter two are the Japonica subspecies (Garris et al., 2005). Significant inter-subspecies differences were detected (P < 0.01; Figures 2A,B). The grain Cd level of Indica rice were significantly higher than that of Japonic (P < 0.01). In contrast, the grain Cd accumulations did not differ by ecotype within a subspecies (P > 0.05). Without considering aro (only three varieties), the ind rice varieties accumulate Cd to levels similar to that of aus (P = 0.31), and tej rice accumulates Cd levels similar to that of trj (P = 0.30). Based on the genetic composition resolution by two-way ANOVA (Table 1), the tej ecotype performed the lowest genetic variations expansions of phenotypic variations (30.6%) and the highest soil-Cd expansions (as 40.0%). These results showed that grain Cd performance of the tej ecotype was the most easily influenced by soil Cd levels and that the ind, aus, and trj ecotypes were insensitive.

We then compared Cd accumulation in the hybrids with the inbred cultivars of the different subspecies or different ecotypes (Figures 2C,D). All the hybrid combinations were phylogenetically grouped into the Indica subspecies in our study (see below). No significant differences in Cd levels were found among the hybrids, ind, and aus ecotypes of the inbred lines (P > 0.05). On the contrary, the tej and trj ecotypes accumulate significantly less Cd than do the ind, aus, or the hybrids (P < 0.01). For most hybrid combinations, the paternal line is usually an elite inbred variety. We therefore also compared the Cd accumulation in the hybrids with their paternal lines (Figure 2E). We found that variation in grain Cd levels between the hybrids and their paternal inbred lines existed, but no uniform pattern was observed. Hybrids accumulate more, less, or similar grain Cd levels to their paternal lines in a case-dependent manner, which agreed with our previous study (Yao et al., 2015; data not shown). In summary, our data indicated the hybrids do not accumulate more Cd than do the inbred cultivars or their paternal inbred lines.

Variation in heading date was present in our hybrid population (Supplemental Figure 1). Rice with a later heading date would be exposed to Cd for a longer time. However, there is no agreement about the relationship between heading date and grain Cd content (Ishikawa et al., 2005, 2010; Liu et al., 2005). To evaluate whether our observation of differences in grain Cd accumulation are determined by heading date, we investigated the relationship between grain Cd concentrations and heading date in hybrid rice. The rice hybrids could be divided into two subgroups: an early maturing group (< 75 days), and a later maturating group (≥75 days), based on the heading date. The average grain Cd level of hybrids in the early maturing group was lower than it was in hybrids in the later maturing group (P < 0.01). Linear regression analysis revealed that there was no significant correlation between grain-Cd with heading date in later maturing varieties (P > 0.05; Supplemental Figure 1B). Although a statistically significant positive correlation was detected for the early-maturing varieties (P < 0.01; R² = 0.21; Supplemental Figure 1A), the low correlation coefficient (R² < 0.3) in linear regression analysis indicated that the positive correlation was meaningless. If both groups were taken into account together, a significant positive correlation, but with R2 of 0.13, between grain Cd concentration and heading date was detected, suggesting the correlation was quite weak (Supplemental Figure 1C). Taking together, we did not detected a strong correlation between grain-Cd concentration and heading date, suggesting that the breeding of rice with low grain Cd levels could not be subjected to the limitation of heading date.

As a nonessential toxic heavy metal, the absorption and translocation of Cd always accompanies other essential elements in plants (Clemens, 2006; Clemens et al., 2013). In this study, by evaluating the relationships between Cd and four other essential elements, Fe, Mn, Cu, and Zn, we did not detect any significant correlation between Cd and any of the four essential metals (P > 0.05; Supplemental Table 2). Thus, the natural variation in grain Cd levels is independent of the accumulation of the other four essential microelements in this study, suggesting that it is possible to breed low grain-Cd hybrid rice without compromising the levels of the other micronutrients.

Because the grain Cd accumulation in the hybrids was found to be similar to that of the Indica inbred cultivars, we determined the genetic backgrounds of the tested hybrids. The hybrids were genotyped with 36 molecular genetic markers covering the 12 chromosomes, including 22 SSR markers and 14 In/Del markers (Supplemental Table 3; Supplemental Data Sheet 1). Based on the genotyping results, 2–10 alleles were detected for each of the markers (an average of 5.17 alleles per marker), which indicated an appropriate level of genetic variation for population analysis. Using the marker genotypes of two typical cultivars, 93-11 and Nipponbare, as the references for the Indica and Japonica subspecies, respectively, a neighbor-joining (NJ) tree was generated using PowerMarker (Ver. 3.25) and MEGA (Ver. 5.0). There were no obvious population differentiations in the hybrid varieties, and all the hybrids grouped into the Indica subspecies (Figure 3A). Additionally, a STRUCTURE analysis produced the highest log-likelihood scores when the population number was set at 14, indicating that most of the hybrids could be classified into 14 sub-groups (Figure 3B). Intriguingly, combined with the geographic data from where the hybrids were developed, an ambiguous geographic distribution could be seen, in which some hybrid varieties with similar geographic origins tended to be clustered into the same clades (Supplemental Figure 2). For example, the hybrids produced from Zhejiang province were mainly clustered into one sub-group, while the hybrids from Fujian province were assigned with those from Jiangxi province. There were also exceptions; for example, hybrids from Hunan province were mainly dispersed among six sub-groups. Nevertheless, with the ambiguous geographic distribution, all the hybrids were still found to belong to the Indica subspecies (Figure 3A). In summary, on the basis of population structure analysis, all of the rice hybrids investigated in this study have a single Indica genetic background.

Many QTLs relating to Cd in rice have been detected in different genetic populations. To determine whether these Cd-related QTLs regulate grain Cd accumulation in rice hybrids, we developed an association analysis of 14 related Cd-QTLs with grain Cd accumulation (Ueno et al., 2009, 2010; Abe et al., 2013; Yan et al., 2013; Zhang et al., 2013; Supplemental Figure 3; Supplemental Table 4). Based on previous reports, a total of 24 markers linked with these QTLs were designed from the Gramene database (http://www.gramene.org/) and used to characterize the genetic polymorphisms of the hybrid varieties. The GLM and the MLM were both used in our association analyses.

In field I, six loci showed significant associations under the GLM model and eight showed significance under the MLM model (P < 0.01). Specifically, the MLM model incorporated both a Q matrix (derived from STRUCTURE) and a Kinship matrix (to correct for the influence of population structure), which could be considered to control false positives. Together, five loci were detected in both models on four chromosomes; Chr 1, Chr 3, Chr 7, and Chr 11, (P < 0.01; Table 2). In field II, eight loci showed significant associations under the GLM model, and five loci were less significant under the MLM model (P < 0.01). Of these, three loci were detected in both models on chromosomes 7 and 11 at P < 0.01 (Table 2), and two significant loci on Chr 3 were also detected at P < 0.05. Taking all associated loci and both fields and models into consideration, a total of four loci were found to be consistently associated with variation in grain Cd levels (Supplemental Figure 3). Two loci, RM8006 on Chr 7 and RM6105 on Chr 11 showed extremely significant associations (P < 0.01), and another two loci, RM132 and RM16153 on chromosome 3, were detected at P < 0.01 in one field condition but at P < 0.05 in the other one.

Of all the four marker loci, the low-Cd alleles have a low distribution in the hybrid population (Supplemental Figure 4). As mentioned above, the early maturating group (< 75 days) accumulated less grain Cd than did the late-maturing group (≥75 days) on average (P < 0.01). Correspondingly, the average low-Cd allelic distribution for the four associated loci in the early maturating group was higher (13.4%) than in the late maturing group (6.8%; Supplemental Figure 5). Therefore, the distribution of the low-Cd alleles from the four associated loci might be a causal factor for the differences in grain Cd concentrations between the two groups. This result could also explain the significant positive correlation between grain Cd concentration and heading date (Supplemental Figure 2), which might be due to the higher distribution of the low Cd alleles in the early maturing group than those in the later maturing group.

In this study, we found that subspecies difference is an important factor with respect to grain Cd concentrations in a population of 68 cultivars (Figure 2A). Although there was no obvious population differentiation detected in the hybrid varieties (Figure 3A), we suspected that the subspecies differentiation of the associated loci might be responsible for the grain-Cd differences in the hybrids. Hence, to test our assumption, we investigated the Indica-Japonica differentiation for the four associated loci based on the RiceVarMap database (http://ricevarmap.ncpgr.cn/), which is a database of rice genomic variations (Zhao et al., 2015). For each associated locus, a total of 45 SNPs randomly distributed across a genomic region of 100 kb covering the associated locus were retrieved from the database. Then, based on the SNPs for each locus, we performed the haplotype network analysis in RiceVarMap (http://ricevarmap.ncpgr.cn/hap_net/; Supplemental Table 5). We identified 10, 12, 13, and 24 haplotype groups that were assembled in accordance with SNPs in the genomic regions flanking loci RM132, RM16153, RM8006, and RM6105, respectively. In support of our assumption, obvious Indica-Japonica differentiation was observed in the genomic regions around RM132, RM8006, and RM6105, but not for RM16153 (Figure 4).

To further confirm the above result, we made a population of 21 hybrid crosses and tested the genetic effects of the four loci in a breeding application. 93-11 and Nipponbare were used to represent the Indica-Japonica diversity. The Indica-Japonica variations were genotyped for RM132, RM8006, and RM6105 in all the 21 hybrid crosses; RM16153 was not used because there are no allelic variations in the 21 hybrid crosses (Figure 5). In total, the hybrids with the Japonica alleles accumulated less Cd than those with the Indica alleles. The Japonica alleles for RM6105 and RM8006 significantly reduced the grain Cd concentration in either the homozygous or heterozygous state (P < 0.01). Because only one line was homozygous for the Japonica allele of RM132, it was omitted from the statistical analysis. Nevertheless, the Indica-Japonica heterozygosis at RM132 was found to reduce the grain Cd concentration significantly (P < 0.01). Although the genetic backgrounds of the 21 hybrid combinations were not clear, the Indica-Japonica diversity at the three loci could reduce the grain Cd concentrations. Thus, the three linked loci, RM132, RM8006, and RM6105, might be applied to MAS for low-Cd hybrid breeding programs. These results confirmed our assumption that genetic diversity, especially the Indica-Japonica differentiation, might play a major role in variation in grain Cd accumulation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.