The maize shoot ionome: Its interaction partners, predictive power, and genetic determinants

Claude Urbany, KWS Saat SE, Grimsehlstr. 31, 37555 Einbeck, Germany. Abstract An improved understanding of how to manipulate the accumulation and enrichment of mineral elements in aboveground plant tissues holds promise for future resource efficient and sustainable crop production. The objectives of this study were to (a) evaluate the influence of Fe regimes on mineral element concentrations and contents in the maize shoot as well as their correlations, (b) examine the predictive ability of physiological and morphological traits of individual genotypes of the IBM population from the concentration of mineral elements, and (c) identify genetic factors influencing the mineral element composition within and across Fe regimes. We evaluated the concentration and content of 12 mineral elements in shoots of the IBM population grown in sufficient and deficient Fe regimes and found for almost all mineral elements a significant (α = 0.05) genotypic variance. Across all mineral elements, the variance of genotype*Fe regime interactions was on average even more pronounced. High prediction abilities indicated that mineral elements are powerful predictors of morphological and physiological traits. Furthermore, our results suggest that ZmHMA2/3 and ZmMOT1 are major players in the natural genetic variation of Cd and Mo concentrations and contents of maize shoots, respectively.


| INTRODUCTION
A better understanding of processes and genes regulating mineral element uptake and how to manipulate the concentrations of mineral elements in aboveground plant tissues holds promise for future resource-efficient and sustainable crop production (Stein et al., 2017) as well as for bio-fortification towards alleviating global malnutrition (http://www.harvestzinc.org/harvestplus). Studies on the composition of plant mineral elements are referred to as ionomics (Lahner et al., 2003). Their aim is to reveal knowledge about the networks controlling uptake, transport, and accumulation of mineral elements.
Ionomic profiling is amenable to high-throughput phenotyping, which when coupled with quantitative genetic approaches such as quantitative trait locus (QTL) mapping or genome wide association studies becomes a powerful tool for gene discovery (Baxter, Gustin, Settles, & Hoekenga, 2013).
The majority of such studies has been performed in model plants (e.g. Bentsink, Yuan, Koornneef, & Vreugdenhil, 2003;Ghandilyan et al., 2009;Klein & Grusak, 2009;Salt, Baxter, & Lahner, 2008;Vreugdenhil, Aarts, Koornneef, Nelissen, & Ernst, 2004;Waters & Grusak, 2008). Only a few studies have performed genetic analyses of the ionome of crop plants such as bean (Blair, Astudillo, Grusak, Graham, & Beebe, 2009), rice (Stangoulis, Huynh, Welch, Choi, & Graham, 2007), Sorghum (Shakoor et al., 2016), or rapeseed (Bus et al., 2014). Two studies used a quantitative genetic approach to dissect the ionome of maize grains using the intermated B73 x Mo17 (IBM) population (Asaro et al., 2016;Baxter et al., 2013). Baxter et al. (2013) described 25 QTL for grain concentrations of nine mineral elements that were detected as significant in two or more locations and, thus, were suggested as targets for biofortification programs. In the study of Asaro et al. (2016), which comprised additional environments to those sampled by Baxter et al. (2013), a total of 79 QTL controlling seed elemental accumulation were detected when each environment was analysed separately. While a set of these QTL was found in multiple environments, the majority were specific to a single environment, illustrating the importance of genotype*environment interactions. Similar results were obtained with segregating populations different from IBM (Gu et al., 2015;Simi c et al., 2012).
However, despite the important role of the mineral elements during the vegetative development of plants, no earlier study examined the inheritance of the shoot ionome of a C4 species.
Most previous studies examined the concentration of mineral elements. Mineral element concentrations reflect primarily the nutritional status of a plant or tissue and provide valuable information on differences in uptake, metabolism, or allocation among genotypes as long as biomass differences between genotypes are low. In our study, we considered additionally mineral element contents, that is, the product of concentration and biomass. These shoot contents reflect the cumulative nutrient uptake and translocation under consideration of the growth response.
The plant's ionomic state results from an interaction between the genotype and the environment, where the latter is strongly influenced by the soil (Baxter et al., 2012). However, our present knowledge of these complex genotype*environment interactions remains sketchy and limited in particular by the vast and discontinuous variation in soil composition at multiple scales in nature (Stein et al., 2017). Joint collections of both plant accessions and adjacent soil were suggested to overcome this bottleneck (Baxter et al., 2012). Apart from being laborious, disadvantages of such approaches are that in most cases the experimental design does not allow to separate genotype*environment interaction variance from error variance. Furthermore, it is impossible to dissect genotype*environment interactions into genotype*individual mineral element interactions.
Therefore, such studies need to be complemented by experiments with artificial growth media, in which nutrient supplies can be controlled. However, such studies are lacking so far.
Iron is one essential mineral element, required for many important biochemical processes, including photosynthesis and respiration, where it participates in electron transport (Marschner, 2012). Therefore, we have chosen Fe exemplarily as the factor to be altered in a hydroponic culture to study a specific form of genotype*environment interaction, the genotype*Fe regime interaction, and to identify the underlying genetic factors.
In agricultural research (Bej & Basak, 2014) and especially plant breeding (Wallace, Rodgers-Melnick, & Buckler, 2018), a high interest exists in exploiting biomolecular signatures to predict complex phenotypes. Baxter et al. (2008) have demonstrated that shoot ionomic signatures in one Arabidopsis thaliana genotype can be used to identify plants that experience an environmental cue. However, to the best of our knowledge, the accuracy of such approaches has not been reported for segregating populations, that is, for entries that differ in their genetic make-up. Furthermore, only few studies have evaluated up to now the predictive ability of mineral elements to estimate genotypic differences in other physiological or molecular traits.
The objectives of this study were to (a) evaluate the influence of

| Culture conditions and measured mineral elements
Maize seeds were sterilized in a 3% NaClO solution for 3 min and then treated with 60 C hot water for 5 min (Benke et al., 2014). Afterwards, seeds were placed between two filter paper sheets moistened with saturated CaSO 4 solution for germination in the dark at room temperature. After 6 days, the germinated seeds were transplanted to a continuously aerated nutrient solution with element concentrations as described by von Wiren, Marschner, and Römheld (1996). The plants were supplied with 100 μM Fe(III)-EDTA for 7 days. From day 14-28, plants were supplied with 300 (Fe sufficient) (Urbany et al., 2013) or 10 (Fe deficient) (Shi et al., 2018) μM Fe(III)-EDTA. The nutrient solution was renewed every third day. Plants were cultivated from day 7 to day 28 in a growth chamber at a relative humidity of 60%, a light intensity of 170 μmol m −2 s −1 in the leaf canopy, and a day-night temperature regime of 16 hr/24 C and 8 hr/22 C, respectively. Four plants of each RIL were grown in one shaded 5 L pot. Two separate pots were used for the two examined Fe concentrations (T = 2). All pots were arranged in a growth chamber following a split-plot design, where the two parental inbreds were included as controls. The entire experiment was replicated R = 3 times. For each RIL, the shoot of the four plants from one pot were pooled so that each RIL was represented by one sample for each of the three replicates in each of the two Fe regimes. Afterwards, the entire sample was ground and the concentrations of 12 mineral elements (Figure 1) were measured using inductively coupled plasma optical emission spectrometry (ICP-OES, iCAP 6,400, Thermo Fisher) according to Shi et al. (2012).
In a previous study, the variation of morphological and physiological traits of the same experiment has been reported (Benke et al., 2014).

| Statistical analyses
Upon the removal of outliers with residuals larger than 2.5 times the standard deviation of the residuals, the concentrations of each mineral element were analysed across both Fe regimes using the following mixed model: where y ijk was the observed elemental concentration of the ith genotype in the pot of the jth Fe regime and kth replication, μ the general mean, g i the effect of the ith genotype, t j the effect of the jth Fe regime, g i *t j the interaction effect of the ith genotype and the jth Fe regime, r k the effect of the kth replication, and e ijk the residual error.
To estimate adjusted entry means (AEM) for all genotypes, g i and t j were considered as fixed and all other effects as random. Furthermore, all effects, except t j , were considered as random to estimate the genotypic variance (σ 2 g ), the interaction variance between genotypes and Fe regime (σ 2 g*t ), and the error variance (σ 2 e ). Statistical significance of σ 2 g and σ 2 g*t has been assessed according to Scheipl, Greven, and Küchenhoff (2008). The broad sense heritability H 2 across both Fe regimes was calculated as: In addition, the concentrations of mineral elements collected from each Fe regime were analysed separately using the following mixed model: where y ik was the observed elemental concentration of the ith genotype in the pot of the kth replication and e ik the residual error. To estimate AEM for all genotypes, g i was considered as fixed and r k as random. Furthermore, g i and r k were considered as random to estimate σ 2 g and σ 2 e . The broad sense heritability H 2 for each Fe regime was calculated in analogy to the method described above: Student's t test was used to examine the statistical significance of the mean difference between deficient and sufficient Fe regimes.
Pairwise Pearson's correlation coefficients were assessed between the 12 mineral elements examined in the current study, 11 morphological and physiological traits evaluated by Benke et al. (2014), and nine mineral elements measured in maize grains by Baxter et al. (2013).
The difference between pairs of correlations was tested using a t-test of the Z-transformed correlations. A principal component (PC) analysis of the normalized and scaled concentrations of 12 mineral elements in the shoot of the maize IBM population as well as the two parental inbred lines measured in sufficient and deficient iron regimes was performed.
All above described statistical analyses were not only performed for the concentration of mineral elements but also for the total content accumulated in the shoots, that is, average content per plant. The average content per plant, which was in the following designated as content was calculated from the product of concentration and the shoot dry weight per plant described by Benke et al. (2014).

| Ionomic and genomic prediction
The prediction of the presence of an environmental perturbation from the physiological response was performed by applying a logistic regression model with the Fe regime as dependent variable. Two sets of mineral elements were used as independent variables: (a) all mineral elements except Fe and (b) all mineral elements except Fe and Zn. We applied fivefold cross-validation (CV) for validation of these predictions (Hjorth, 1994). For this purpose, the 85 RIL were randomly sub- GBLUP method was used as implemented in the R package sommer (Covarrubias-Pazaran, 2016). Fivefold CV was applied, as described above. We calculated the predictive ability [r(y pred , y obs )] as the Pearson correlation between the observed AEM and the predicted genotypic values.

| QTL analyses
The publicly available genotypic data for the RIL as well as the genetic map positions of these markers on the IBM2 map were the basis of our analyses. Markers that showed a highly significant (p < .001) distorted segregation were excluded. The remaining 1,212 markers were used for the QTL analyses. QTL analyses to detect genome regions linked to the difference in the concentration and content of mineral elements across the two examined Fe regimes were conducted with Rqtl (Broman, Wu, Sen, & Churchill, 2003). Standard interval mapping using the Haley-Knott regression algorithm (Haley & Knott, 1992) and a model that included the Fe regime as additive covariate were used in a first step.
Forward selection was used in the following to determine the maximum number of QTL to include in the model selection procedure.
An automated forward and backward search algorithm was then applied to perform multiple QTL mapping using a model that included again the Fe regime as additive covariate. Model selection was based on the highest penalized LOD score with penalties determined through 1,000 permutations. In the next step, the presence of epistatic interaction among all pairs of QTL with main effects were examined. Afterwards, a forward and backward search algorithm was applied to add significant QTL*Fe regime interactions to the model, where significance of all models tested was empirically determined using 1,000 permutations.
Confidence intervals for QTL were calculated using a 1.5 LOD drop technique (Manichaikul, Dupuis, Sen, & Broman, 2006). In order to obtain unbiased estimates of the proportion of the explained phenotypic variance, fivefold CV accounting for genotypic sampling as applied as described earlier for QTL mapping (Utz, Melchinger, & Schön, 2000) and the mean proportion of the phenotypic variance explained by one or multiple QTL (R CV ) was calculated.

| Characterization of genes in QTL confidence intervals
The molecular markers flanking the QTL confidence interval were identified and their physical position on the B73 AGPv4 sequence was derived. For those flanking markers for which a physical map position could not be derived, it was estimated based on the nearest locus on the IBM2 2008 Neighbours map for which a physical positioning was available. The genes in the physical QTL confidence interval were extracted from the B73 AGPv4 sequence. The polymorphisms within the physical confidence interval were extracted from the maize haplotype version 3 (Bukowski et al., 2018) after converting the assembly to AGPv4 using CrossMap (Zhao et al., 2014). The annotation of these polymorphisms was performed using the Ensembl Vari- After converting the AGPv4 based positions of candidate genes to AGPv2 based, the v2.7 data from genotyping by sequencing experiments of maize inbreds of the USA national seed bank (Romay et al., 2013) was used to extract existing haplotypes based on polymorphism information from the candidate genes. These 1,749 inbreds were considered in our study for which the information of the US state of origin was available in the Germplasm Resources Information Network database. For haplotypes occurring in the entire sample with a frequency >2.5%, a haplotype network was built using an infinite site model (Hamming, 1950).
If not stated differently, all statistical analyses were performed using statistical software R (R Development Core Team, 2016).

| RESULTS
We characterized the variability of the shoot concentrations of mineral elements across two Fe regimes using mixed model analyses to quantify the influence of the various sources of variance. For the concentrations of all mineral elements examined in our study, a significant (α = 0.05) genotypic variance (σ 2 g ) was observed except for K and Zn ( Figure 1). Similarly, σ 2 g was significant (α = 0.05) for the content of all mineral elements except for Cu and Mn. The variance of genotype*Fe regime interactions (σ 2 g*t ) was on average across the concentrations of all mineral elements even more important than σ 2 g and was significantly different from 0 (α = 0.05) for all mineral elements. In addition, σ 2 g*t of the concentrations of mineral elements was considerably more pronounced compared to that of the contents of mineral elements. As a consequence, the broad sense heritabilities (H 2 ) of the concentrations of mineral elements for the sufficient and deficient Fe regime were with a range from 0.39 to 0.95 slightly higher than H 2 across both Fe regimes, whereas the opposite trend was observed for the mineral element contents.
The AEM for all mineral element concentrations except Fe was on average across all RIL significantly higher (α = 0.05) under the deficient Fe regime compared to the sufficient regime, while the opposite was true for Fe ( Figure 2). In contrast, the AEM for the content of all mineral elements except Cd, Cu, Mn, and Zn was on average across all RIL significantly higher (α = 0.05) under the sufficient Fe regime compared to the deficient regime, while the opposite was true for Zn. A strong transgressive segregation was observed in the RIL for the concentrations of all mineral elements which suggests that both parental inbreds contribute alleles to the progenies that increase the concentrations of the studied mineral elements. For the mineral element contents, transgressive segregation was less pronounced compared to that observed for the concentrations of mineral elements.

F I G U R E 1
Analysis of variance for shoot concentrations and contents of 12 mineral elements. Linear mixed models were used to describe (a) the variance contributed from genotype (σ 2 g ), genotype*iron (Fe) regime interaction (σ 2 g*t ), and error (σ 2 e ). σ 2 g and σ 2 g*t estimates marked by *, **, and *** are significantly (α = 0.05, 0.01, and 0.001) different from 0. (b) Estimates for heritability (H 2 ) across as well as for the individual Fe regimes were calculated In the PC analysis of the concentrations of mineral elements, the first two principal components explained 19.2 and 13.9% of the variance ( Figure S1), where no obvious clustering of the RIL was observed with respect to these two principal components. The same trend was observed in the PC analysis of the mineral element contents (data not F I G U R E 2 Frequency distribution of the adjusted entry means (AEM) of the (a) concentrations, (b) shoot contents of 12 mineral elements in the maize IBM population measured at sufficient (white) and deficient (grey) iron regimes. The AEM of the parental inbreds B73 and Mo17 are represented by a square and a triangle, respectively. The significance level of a t-test examining the difference of one mineral element examined in two different iron (Fe) regimes across the population are given above the horizontal line. Pearson's correlation coefficient (ρ) was calculated between the AEM of both Fe regimes. *, **, ***: p = .05, .01, and .001, respectively; ns, not significant regardless of whether the concentration or the content of mineral elements was considered.
The pair of mineral element concentrations with the highest correlation coefficient that was consistent across both Fe regimes was P/Mn, while that with the lowest was Mo/Fe. In total, we observed for 10 pairwise correlations of mineral element concentrations that did not involve Fe significant (α = 0.05) differences between the Fe sufficient and the Fe deficient regime (Figure 4a).
The same general trends that were observed for the concentrations of mineral elements were also observed for the contents of mineral elements (Figure 3b). However, the relationship between the correlation coefficients calculated between the contents of two mineral elements under the Fe sufficient and the Fe deficient regime was more tight than that observed for the concentrations of mineral We examined the correlations between the mineral element concentrations in maize shoots and morphological and physiological traits of the same plants (Benke et al., 2014). In the sufficient Fe regime, we observed tight correlations between the shoot concentration of P and root weight (RW), shoot dry weight (SDW), and lateral root formation (LAT), whereas under deficient Fe conditions, the tightest correlations were observed between P, Mn, and Zn concentrations and the shoot dry weight and different leaf greenness parameters (SP) ( Figure S3).
These correlations were in close agreement with earlier studies and confirmed here across a RIL population.
From the ionomic profile of maize shoots, we evaluated the possibility to predict an environmental perturbation. This was done by predicting the Fe regime in which the RIL of the IBM population was cultivated based on the measured concentrations of mineral elements.
When using all mineral elements except Fe and Zn as independent variables, the cross-validated predictive ability was very high with a value of 0.90. This ability increased to 1.0, when Zn was included as independent variable. In a second approach, we examined here the prediction of physiological and morphological traits assessed previously by Benke et al. (2014) in the IBM population from molecular marker data. The cross-validated predictive ability ranged from 0.13 to 0.53 for the sufficient Fe regime (Table 1). For the deficient Fe regime, the observed predictive abilities were higher and ranged from 0.23 to 0.57. The same trend was observed when using the mineral elements as predictors. However, the such observed predictive abilities were for the sufficient Fe regime for five physiological and morphological traits significantly (α = 0.05) higher than for predictions made from molecular marker data. For the deficient Fe regime, this trend was observed for nine of the 11 examined traits.
To unravel the genetic determinants of the shoot ionome of maize, a linkage mapping approach was used. For the concentration of all mineral elements except K, genome regions with significant effects were identified (Table S1) In general, we observed a clustering of genome regions that

| Natural variation in the mineral element profile of maize shoots and the interaction with the Fe regime
Across the RIL of the IBM population, we observed a high phenotypic variability in the concentration but also the content of all mineral elements ( Figure 2). The performed mixed-model analyses revealed that for most mineral elements concentrations as well as contents this variation was due to significant (α = 0.05) genotypic (σ 2 g ) variance ( Figure 1). These findings indicate that the examined population is appropriate to be used to identify the genetic factors that contribute to the variation of mineral element concentrations in maize shoots.
This holds especially true as we have studied an intermated RIL population, in which additional rounds of intermating cause a rapid decay of linkage disequilibrium. Thus, correlations between mineral elements observed in our study are less likely caused by linkage but are rather due to pleiotropy. Therefore, the examined population is also particularly suitable to validate correlations among mineral elements.
Due to the design of our study that is, the assessment of the shoot ionome of the IBM population in two contrasting Fe regimes, we were not only able to validate correlations between pairs of mineral elements but to compare the correlation of two mineral elements across different Fe regimes. By that, we identified interactions among Fe and two other mineral elements. We observed significant Earlier studies have reported the interaction of Fe with the concentration of several mineral elements, mostly with S, P or other metals (Astolfi, Zuchi, Passera, & Cesco, 2003;Kanai, Hirai, Yoshiba, Tadano, & Higuchi, 2009;Mori & Nishizawa, 1987;Zheng et al., 2009). Under Fe deficiency, all of these correlations turned more negative (Figure 4a), which is not only related to endodermal suberization but also to the Fe deficiency-induced expression of genes involved in Fe acquisition which favour also Zn and Mn acquisition (Eroglu, Meier, Wiren, & Peiter, 2016;Leskova, Giehl, Hartmann, Fargasova, & von Wiren, 2017). In maize, this includes the release of phytosiderophores and uptake of phytosiderophore-chelated metals, such as Zn, Cu and Ni (Schaaf et al., 2004;von Wiren et al., 1996).
In addition, we observed a particularly strong difference between correlations were negative (Figure 4a). This link between P and Fe is in accordance with earlier studies. DeKock, Hall, and Inkson (1979) and Zheng et al. (2009) have shown that elevated P supplies decrease the availability and mobility of Fe in soils as well as in plants and therewith provoke Fe deficiency-induced chlorosis (Ajakaiye, 1979).
The results of Shi et al. (2018) indicate that Fe retention in the hemicellulose fraction of roots is an important determinant of the tolerance to Fe deficiency-induced chlorosis of graminaceous plant species with low phytosiderophore release, like maize. Then, P-dependent Fe retention in the hemicellulose fraction may become a determining factor for Fe provision of the shoot and, thus, for chlorosis susceptibility.
With the increasing knowledge on such mechanistic processes underlying nutrient correlations, the shoot ionome becomes more and more informative (Huang & Salt, 2016) and paves the way to identify and assign anatomical changes or physiological processes to allelic variation. Baxter et al. (2008) showed that the impact of an environmental perturbation on a plant can be predicted by changes in its ionome. While this principle was demonstrated for the Col-0 accession in A. thaliana when subjected to Fe deficiency, we first addressed the question whether the shoot ionome is also a reliable predictor for the nutritional status in a RIL population subjected to the same nutritional perturbation. Using the IBM population for this purpose is especially meaningful, because it has been generated from a cross between two parental lines opposing in P and Fe efficiency (Benke et al., 2014;Kaeppler et al., 2000;Shi et al., 2018). Indeed, our analysis revealed a  (Campos et al., 2017). These findings indicate that even across diverse genetic materials, the profile of the mineral elements is a reliable proxy for the physiological state and allows identifying nutritional perturbations.

| Predictive power of the shoot ionome
We then addressed the question whether the shoot ionome is also a reliable predictor for morphological and physiological traits perturbed by an altered Fe regime. In fact, when predicting from the concentration of mineral elements, physiological and morphological traits of individual RIL (cf. Benke et al., 2014), cross-validated predictive abilities up to 0.77 were observed (Table 1). An increasing prediction ability under Fe deficiency indicated that an imposed nutritional disorder even increases the predictive power of the ionome.
We then compared the prediction of physiological and morphological traits by mineral elements with that by molecular markers.
Unexpectedly, ionome-based predictive abilities were considerably higher compared to the predictive abilities based on molecular marker profiles using GBLUP ( In this study, we exploited the significant genotypic and genotype*Fe regime interaction variance found in the IBM population ( Figure 1a) and employed quantitative genetic analyses in order to dissect this variation in the underlying genetic factors. For all mineral elements examined in the frame of this study, except for the concentration of K and the content of P, genomic regions were identified that explained a significant part of the genotypic or genotype*Fe regime interactions (Table S1 and S2). The proportion of variance explained for one mineral element by all QTL interacting with the environment simultaneously was considerably lower than that of the main effect QTL. This finding suggests that albeit the variance of genotype*Fe regime interactions for the concentration of mineral elements was considerably higher than the genotypic variance ( Figure 1) the QTL interacting with the environment were not detected in our study. One major reason might be that each QTL explains only a small proportion of the genotype*Fe interaction.
Due to the use of the IBM population, which has compared to non intermated linkage mapping populations an increased mapping resolution (Lee et al., 2002), the confidence intervals observed in our study were small for most of the 60 QTL (Table S1 and S2). Out of the detected QTL, four QTL had overlapping confidence intervals for the concentrations and contents of the same mineral element. Two of these pairs of QTL jutted out regarding the LOD score as well the proportion of the explained phenotypic variance and these will be discussed in more detail.

| C-Cd1 and A-Cd1
We did not add Cd to the nutrient solution in which the IBM population was cultivated. Nevertheless, we observed under both Fe regimes low Cd concentrations in maize shoots. To rule out the effect of Cd translocation from the seeds to the growing shoot tissue, we evaluated the Cd concentrations in the grains of a subset of the IBM population. The overall amount of Cd in the grain made up less than 4% of the shoot content, that is, the total amount of Cd accumulated in the above-ground biomass. This indicated that the highly heritable pattern of Cd concentrations that we observed (Figure 1) was caused by Cd traces in the chemicals or the technical devices used to prepare the nutrient solution. As therefore all RIL were exposed systematically to the same Cd concentration in the nutrient solution, we interpreted the observed variation in our study.
We detected in our study two overlapping QTL on chromosome 2 for Cd concentration and content. These QTL were in very good accordance with the QTL detected for Cd content in maize kernels observed in a B73*IL13H population by Baxter et al. (2014). This observation indicates that the mechanisms contributing to the accumulation and sequestration of this mineral element in maize shoots and kernels are similar. However, for nine other mineral elements, correlation coefficients between the mineral element concentrations in maize shoots from our study and in maize kernels from Baxter et al. (2013) were low (data not shown). Hence, a common regulation across tissues exists for some mineral elements but not for all.
Across both Fe regimes, the trait-decreasing allele was provided by B73 ( Figure 6). This indicated that inbreds and hybrids carrying the B73 allele are suitable to produce maize on soils with increased Cd accumulations without increasing the Cd concentration in the harvest product.
Within the confidence interval of C-Cd1/A-Cd1 two genes, Zm00001d005189 and Zm00001d005190, were located directly adjacent to each other that are annotated as HMA genes. In the coding sequence of both genes, non-synonymous polymorphisms have been identified that are predicted (McLaren et al., 2016) to have moderate or even high variant consequence (Figure 7). Zm00001d005189 is based on its amino acid sequence not of particular similarity to any of the HMA genes of Arabidopsis or rice (data not shown). In contrast, Zm00001d005190 has a high homology to OsHMA3 indicating that Zm00001d005190 could be the ortholog ZmHMA3 which is in accordance with results of Zhao et al. (2018).
Our result indicated that the Mo17 allele at Zm00001d005190, which is presumably non-functional, is present at medium frequency in the maize germplasm of the USDA genebank. This suggests that under regular field conditions the higher accumulation of Cd does not have a negative agronomic side effect. Another explanation is that commercially used maize hybrids are typically crosses between B73-type and

| C-Mo1 and A-Mo1
In the confidence interval of these QTL, a homolog of A. thaliana's Mo transporter (MOT1) (Tomatsu et al., 2007) was detected. We observed between the coding sequences of B73 and Mo17 one nonsynonymous sequence variant (Figure 7) that is predicted to have only a medium effect on the protein. Instead, in accordance with findings of Baxter et al. (2008) for A. thaliana, we consider differences in gene expression as more relevant, because in the deficient Fe regime mRNA levels of Zm00001d033053 were about 1,000-fold higher in Mo17 than in B73, whereas in the sufficient Fe regime this difference was only sixfold (Urbany et al., 2013). This Fe dependence of the Mo accumulation was not only observed at the gene expression level but also at the QTL level. C-Mo1 showed the strongest QTL*Fe regime interaction of this study. However, further research is required to unravel the molecular mechanism of the interaction between Fe and Mo.

| CONCLUSIONS
Our results suggest that the shoot ionome is under tight genetic control, and that strong interactions of a genotype with the environment, in our case the Fe regime, are driving the variation for the concentration as well as the content of mineral elements. Furthermore, the high predictive abilities found in this study indicate that mineral elements are powerful predictors of morphological and physiological traits.
Their use for the prediction of further, for example, yield-related traits, merits to be examined in further studies. We propose that ZmHMA2/3 and ZmMOT1 are major players in the natural genetic variation of Cd and Mo concentrations and contents in maize shoots.
Furthermore, our results indicate that even in elite breeding material non-functional alleles segregate at major genes for uptake and translocation of different mineral elements.